FR2953675A1 - Method for controlling transfer of video sequence transmitted over set of broadcasting sessions in communication network, involves adapting subscription to sub assembly of broadcasting sessions based on importance of loss of data - Google Patents

Abstract

The method involves detecting loss of data affecting video data of set of images, and selecting decoding technique based on localization of video data affected by loss of data in set of images. A length of another set of images greater than the length of the former set of images is detected, and an importance of loss of data affecting video data of set of images is evaluated. A subscription is adapted to the sub assembly of broadcasting sessions based on the importance of loss of data. Independent claims are also included for the following: (1) a device for controlling transfer of video sequence comprising an identification unit (2) a computer readable information storage unit comprising a method for controlling transfer of video sequence transmitted over set of broadcasting sessions in a communication network (3) a computer program product comprising sequences of instructions for a method for controlling transfer of video sequence transmitted over set of broadcasting sessions in a communication network.

Description

The present invention relates to a method of control by a client device of the transfer of a video sequence. It also relates to a control device implemented by a client device during the transfer of a video sequence.

In general, the invention relates to the transmission of a video sequence between a server and at least one client device over an unreliable communication network, that is to say in which the transmission conditions are not stable and / or data loss may occur.

Throughout the following description, the term "loss" refers to both missing data and erroneous data detected in the video data of the transmitted video sequence. The invention is more particularly concerned with the regulation by a client device of the bit rate of the video data intended for it, based on information on the detection and localization of video data affected by losses. One places oneself in the case where the coding of the video sequence is of hierarchical or "scalable" type, that is to say that it implements one or several hierarchical levels, also called levels or layers of scalability. A non-limiting example of such a scalable encoding format is SVC ("Scalable Video Coding"). Three types of scalability have been defined in the SVC standard: spatial, temporal and quality scalability, quality scalability also being known as SNR (Signal to Noise Ratio) scalability, signal-to-noise ratio. . The temporal scalability makes it possible to modify the temporal resolution of a sequence (that is to say the number of images per second represented by the coded data) by suppressing certain images, this deletion taking into account the possible dependencies between the images. .

Spatial scalability involves inserting several spatial resolutions (corresponding to different numbers of pixels represented by the encoded data) into a video stream, the lowest resolution being used for prediction of higher resolutions. Scalability in quality takes three different forms: Coarse Grain Scalability (CGS), Medium Grain Scalability (MGS) and Fine Grain Scalability (FGS). "Fine Grain Scalability"). CGS scalability uses the same concepts as spatial scalability, except that for CGS, over-sampling operations of inter-layer prediction are omitted. The FGS scalability allows the creation of a bitstream that can be truncated at any point while remaining decodable. MGS scalability has been defined as intermediate between CGS and FGS: it offers decoding points in the bitstream finer than CGS but does not allow truncation at any point like FGS. MGS scalability is often considered to provide sufficient granularity for realistic network conditions. The transmission of a video on a network is facilitated by the introduction of the concept of NAL entity (in English "Network Abstraction Layer"). A NAL is an elementary bit-stream transfer unit that provides in its header elements for describing the data carried in the data portion of the NAL. By way of non-limiting example, for the transmission of an SVC stream, the NAL can be seen as a video data container (in English). the NAL header includes in particular the following parameters: • Dependency_id (DId): specifies the spatial layer of the video data contained in the NAL; • Quality_id (QId): specifies the quality layer of the video data contained in the NAL; • Temporal_id (TId): Specifies the temporal layer of the video data contained in the NAL. All the NALs corresponding to the same coded picture form an entity called AU (in English "Access Unit").

In the context of a video broadcast application (often called "video streaming" in English) between a server system and a client device over a network, the video data is generally transported in RTP packets (transport protocol). real-time Transport Protocol) because RTP offers an interesting set of error-resistance tools The RTP packet header contains in particular two pieces of information: • a sequence number, and • a timestamp element or Timestamp (Timestamp): The sequence number is chosen randomly for the first packet of a video broadcast session, and then increments by 1 each time a new package is created for that application. The sequence number provides a way of detecting that a loss has occurred in a set of packets, because RTP prevents sequence number hops if a client detects in the numbers. sequencing a jump greater than 1 between two packets received consecutively, it can deduce that at least one loss has occurred. The time stamp (in English "timestamp") represents the time of creation of the transported data. For example, all SVC data corresponding to the same instant must bear the same time stamp if they are carried by several packets. All of these parameters make it possible to track the transmission of packets carrying coded video data over a transmission network. Indeed, communication networks using packet transmission are regularly confronted with data loss problems. Video applications are particularly sensitive to these losses since real-time decoding prevents the retransmission of lost data and the errors propagate in the received data due to the conventional predictive coding methods implemented. The errors or losses of data detected in the communication networks can have several causes such as packet collisions, interference problems, or are due to congestion of the communication network. The video data transmission is then faced with isolated losses, when the events generating these losses are of short duration (interference, packet collision) or losses occurring over a longer period, when they are due for example to congestion of the communication network. When a communication network carries scalable video data, each hierarchical level of the encoded video data is carried in a dedicated multicast session. The client devices of this application can subscribe (in English "to subscribe") to a subset of these multicast sessions, thus receiving different levels of hierarchy according to their decoding capacity.

In such a case, it is particularly advantageous to be able to evaluate the extent to which a hierarchy level is affected by losses, in order to avoid routing hierarchy levels to a client device that will not provide sufficient improvement in the hierarchy. spatial, SNR or temporal quality when decoding the video sequence.

Thus known error detection techniques implemented at a client device of a communication network for detecting errors and locate the image affected this error. US-A-7,072,403 proposes a cooperation between the network layer and the decoder in a system where the packets carry blocks of data corresponding to a set of pixels. The client device detects errors in the packets at the network level by using error detection codes and provides the corresponding information to the decoder. The decoder can thus decide whether the blocks contained in the packets are actually affected by a loss.

It is then possible to decide not to decode in the received video data stream the hierarchical level whose video data is significantly affected by data loss. Since a hierarchical level is not decoded by the client device, it is advantageous to propose a method for avoiding the transmission of this hierarchical level to the client device in the context of a multicast session transmission. FIGS. 1 to 3 show a conventional control by a client device of the transfer of a video sequence when isolated losses occur on the communication network. Here, an S server transmits a video sequence as hierarchically coded video data to a set of six client devices C1, C2, C3, C4, C5, C6. The data stream is thus scalable and transmitted over three multicast sessions, each hierarchical level being transmitted on the communication network respectively by a multicast session. The dashed arrows represent a first multicast session carrying the lower hierarchical level. The dashed arrows represent a second multicast session carrying a second hierarchical level. The solid arrows represent a third multicast session carrying a third hierarchical level, superior to the previous ones. The video sequence is transmitted by three routers R1, R2, R3.

The first two routers R1, R2 are high speed routers whereas the router R3 is a low rate router. Thus, the client device C1, connected only to the first two routers R1, R2 receives a high data rate and can thus receive the three multicast sessions.

On the other hand, the client devices C2 to C6, connected to the third router R3, are only able to receive the first two multicast sessions of the video sequence.

In such a case, as illustrated in FIG. 2, when a client device C2 tries to subscribe to the third multicast session, this subscription causes congestion at the level of the communication network, and in particular the third router. R3.

However, this congestion affects the transfer of video data to all client devices C2 to C6 connected to the third router R3. FIG. 3 illustrates the conventional reaction of the client devices C2 to C6 during the detection of data losses due to network congestion at the level of the router R3. In such a case, when data losses are detected by the client devices C2 to C6, they unsubscribe from the higher level multicast session. Thus, in the example illustrated in FIG. 3, the client device C2 at the origin of the congestion unsubscribes from the third multicast session, while the client devices C3 to C6 unsubscribe from the second multicast session . This conventional method of controlling the transfer of the video sequence thus has the disadvantage of causing unwanted reactions at the client devices of the communication network. In fact, in the previous example, the congestion caused by the client device C2 was temporary and isolated and should not have affected the subscription of the client devices C3 to C6 to the multicast sessions. A solution proposed by S.MAC CANNE and V. JACOBSON, ACM SIGCOMM 96, called "Receiver Layered Multicast" or RLM has been proposed to avoid untimely reactions on a communication network. In principle, the RLM solution provides that, when a customer tries to subscribe to an additional multicast session, it informs other clients of the communication network by sending a multicast message ("multicast message").

The other client devices are thus informed that one of the client devices is trying to subscribe to an additional session and that this attempt may cause temporal congestion in the communication network. If such congestion takes place immediately after the receipt of such a message by the client devices, the latter do not react to the detection of data loss affecting the video data received, and in particular, do not change their subscription to the subsystem. set of multicast sessions. However, this solution requires the creation of a protocol allowing the client to inform other clients by a multicast message. Moreover, this solution is adapted only to this type of congestion and is not adapted to manage temporary congestion created by other applications. FIG. 4 illustrates the behavior of a client device implementing the RLM solution for controlling the transfer of a video sequence transmitted by a subset of multicast sessions. During the first three periods of time A, the client device subscribes successively to three first multicast sessions. Since there is no congestion, in the next period it attempts to subscribe to an additional multicast session. However, as illustrated in FIG. 4, a congestion of the communication network, detected from the loss of data affecting the video data received, appears as soon as the cumulative bit rate of the four multicast sessions is greater than the bit rate available on the network. Communication. When data loss is detected, the client device unsubscribes from the higher level multicast session. Then, the client device tries to subscribe again to this fourth multicast session, but the time period between each subscription attempt is increased, for example doubled.

The increase of this period of time avoids causing too many and too frequent congestions on the communication network. As mentioned above, this RLM solution has the disadvantage of causing the top-level multicast session always to be unsubscribed when data loss is detected. When the congestion is temporary, the use of the transmission possibilities of the communication network is not optimized.

Furthermore, as well illustrated in Figure 4, by increasing the time period between each subscription trial, the system loses responsiveness. It prevents the client device from reacting effectively when the throughput of the communication network increases. For example, as illustrated in FIG. 4, when the period of time is equal to, for example, eight A, an increase in the bandwidth of the network at the beginning of this period will not allow a subscription to the fourth multicast session before the end of this period while the available bandwidth would allow an enhanced subscription by the client device to the different available multicast sessions.

The present invention is intended in particular to avoid an unwanted reaction of a client device in the management of its subscription to multicast sessions for the transfer of a video sequence transmitted over a communication network, without the need for communication between the client devices of the communication network.

For this purpose, the present invention relates, according to a first aspect, to a method of control by a client device of the transfer of a video sequence transmitted on a set of multicast sessions by a server in a communication network, the video sequence being coded according to a hierarchical video coding on a set of hierarchical levels, each hierarchical level being transmitted on the communication network respectively by a multicast session, the client device subscribing to at least a subset of multicast sessions.

The control method comprises the following steps: detecting, on a first set of images of the video sequence, data losses affecting video data of the first set of images; selecting a decoding technique according to the location of the video data affected by the data losses in the first set of images; detecting, on a second set of images of the video sequence, the length of the second set of images being greater than the length of the first set of images, data losses affecting video data of the second set of images; - evaluating the importance of the data losses affecting the video data of the second set of images; and - adapting the subscription by the client device to a subset of multicast sessions based on the amount of data loss affecting the video data of the second set of images. The control method according to the invention thus makes it possible to choose the subset of multicast sessions to which the client device subscribes, taking into account the losses of data over a period of time greater than that used to process the losses at the time. level of decoding, in particular to avoid unsubscribing to a multicast session in the event of data loss due to temporary congestion of the communication network. Thus, a short-term error detection step on a first set of images is used to determine which hierarchical levels must be decoded, while a long-term error detection step on a second set of images is used to determine the subset of multicast sessions to which the client device subscribes.

The control method thus makes it possible to adapt the reaction of the client device in the control of the transmission of a video sequence depending on the type of congestion, long-term or transient, on the communication network. Moreover, the control method can be implemented by each client device without requiring information on the behavior of other clients and in particular the subscriptions or unsubscriptions of other client devices to a set of multicast sessions on the communication network. . The control method according to the invention thus makes it possible to adapt the subscription of the client device in an optimized manner on the communication network, whatever the nature and origin of the congestions appearing on the communication network. According to a practical embodiment of the invention, the step of selecting a decoding technique is adapted to select at least one hierarchical level not to be decoded.

In practice, at the evaluation stage, the importance of data loss is evaluated on video data of a given hierarchical level and at the adaptation stage, the client device unsubscribes from a broadcast session. group conveying the given hierarchy level when the number of evaluated data losses exceeds a predetermined threshold value of data loss. Preferably, the predetermined threshold value has a value dependent on the given hierarchical level. For example, the predetermined threshold value is even higher than the given hierarchical level is low.

According to an advantageous characteristic, in the adaptation step, the client device also unsubscribes from at least one multicast session transmitting video data of hierarchical level higher than the given hierarchical level. Thus, when large congestions occur on the communication network, it is possible to unsubscribe the client device to multiple multicast sessions simultaneously, thereby speeding up the adaptation of the transmission of the video sequence. function of the available flow on the communication network. This can result in unsubscribing at one or more multicast sessions transmitting video data at a higher hierarchical level than the hierarchical level affected by data loss, especially when the video data of the higher hierarchical level depends on the video data of the lower hierarchical level. affected by data loss. Advantageously, the video data received by a client device are decodable to the extent that they are not dependent on a lower hierarchy affected by losses. In practice, a hierarchical level can be characterized by three indices respectively representing a level of spatial hierarchy, a level of temporal hierarchy and a hierarchy level in quality, and a first hierarchical level is greater than a second hierarchical level when at least one representative indices of the first level have a value greater than the value of the same representative index of the second level. According to one embodiment of the invention, the control method comprises the following steps when the server inserts priority levels respectively into video data packets respectively transferred on a set of multicast sessions, the priority level inserted in a video data packet being dependent on the hierarchical level of the video data of the packet: identification of the priority level associated with a video data packet affected by the loss of data detected during the detection step on a second set of data images of the video sequence; and - adapting the subscription by the client device to the subset of multicast sessions according to the identified priority level. According to an advantageous characteristic of the invention, the control method comprises a step of varying the length of said second set of images according to the adaptation of the subscription by the client device to the subset of broadcast sessions. grouped at said adaptation step. In practice, at said step of variation, the length of the second set of images is decreased when at said adaptation step, the client device unsubscribes from at least one multicast session. Alternatively, at said step of variation, the length of the second set of images is increased when at said adaptation step, the client device maintains its subscription to the subset of multicast sessions.

In practice, the first set of images is a group of images, the second set of images consisting of a set of successive image groups including the first set of images. In practice, the length of the second set of images is increased or decreased by the length of a group of images.

According to another preferred feature of the invention, when the length of the second subset of images is equal to a minimum prefixed value, a time interval between two subscription attempts by said client device to an additional multicast session , in addition to said subset of multicast sessions, is increased.

This avoids multiplying the subscription attempts to an additional multicast session when this subscription systematically causes congestion on the communication network. According to a second aspect, the present invention relates to a method for decoding by a client device a video sequence transmitted on a set of multicast sessions by a server in a communication network, said video sequence being encoded according to a hierarchical video coding on a set of hierarchical levels, each hierarchical level being transmitted over the communication network respectively by a multicast session, said client device subscribing to at least a subset of multicast sessions, the decoding method comprising the following steps: - detecting, on a first set of images of the video sequence, data losses affecting video data of the first set of images; decoding a hierarchical level selection according to the location of the video data affected by the data losses in the first set of images; detection, on a second set of images of the video sequence, the length of the second set of images being greater than the length of the first set of images, data losses affecting video data of the second set of images ; - evaluating the importance of the data losses affecting the video data of the second set of images; and - adapting the subscription by the client device to the subset of multicast sessions based on the amount of data loss affecting the video data of the second set of images. Correlatively, the present invention also relates to a device for controlling the transfer of a video sequence transmitted on a set of multicast sessions by a server in a communication network, said video sequence being coded according to a hierarchical video coding on a set hierarchical levels, each hierarchical level being transmitted over the communication network respectively by a multicast session, said client device subscribing to at least a subset of multicast sessions, the control device comprising: means for detecting, on a first set of images of the video sequence, data losses affecting video data of the first set of images; means for selecting a decoding technique according to the location of the video data affected by the data losses in the first set of images; detection means, on a second set of images of the video sequence, the length of the second set of images being greater than the length of the first set of images, data losses affecting video data of the second set of images; 'images; means for evaluating the importance of the loss of data affecting the video data of the second set of images; and means for adapting the subscription by the client device to the subset of multicast sessions as a function of the importance of the data losses affecting the video data of the second set of images. This control device has features and advantages similar to those previously described in connection with the control method. The present invention also proposes a client device in a communication network, comprising calculation means, storage means and means of communication with the network, the calculation means co-operating with the storage means and the means of communication with the network. the network for implementing the control method and / or the decoding method according to the invention. It also relates to a means for storing information readable by a computer or a microprocessor retaining instructions of a computer program adapted to implement the control method and / or the decoding method such as those briefly described below. above. Finally, the present invention is also directed to a computer program product that can be loaded into a programmable apparatus, including instruction sequences for implementing a control method and / or a decoding method as briefly described above. when this program is loaded into and executed by the programmable device. The particular features and advantages of the client device, the information storage means and the computer program product are similar to those of the methods briefly described above. These features and advantages are therefore not repeated here. Other features and advantages of the invention will become apparent in the description below.

In the accompanying drawings, given by way of non-limiting examples: FIGS. 1 to 3 illustrate the principle of controlling the transfer of a video sequence implemented in the prior art over a communication network; FIG. 4 is a diagram illustrating the adaptation of a subscription of a client device in the prior art as a function of the bit rate available on a communication network; - Figure 5 schematically shows a particular embodiment of a video transmission system adapted to implement the present invention; FIG. 6 is a flowchart illustrating the main steps of the control method according to a particular embodiment of the present invention; FIGS. 7 and 8 are flowcharts illustrating in greater detail, in a particular embodiment, the loss detection step illustrated in FIG. 6; FIGS. 9 and 10 are flowcharts illustrating, according to a particular embodiment, the step of selecting the hierarchical levels to be decoded illustrated in FIG. 6; FIG. 11 is a flowchart illustrating, according to a particular embodiment, the step of evaluating the importance of the losses of data illustrated in FIG. 6; and FIGS. 12 and 13 are diagrams illustrating the adaptation of a subscription of a client device embodying the present invention, as a function of the available data rate on a communication network. In the remainder of the description, a video sequence transmitted between a server of a communication network and one or more clients on this communication network is considered. The video sequence is transmitted in a continuous stream by a set of multicast sessions addressed by a server to a client device of the communication network.

The present invention aims to control the subscription of the client device to a subset of multicast sessions in order to optimize the decoding of the video sequence, in particular by improving the reaction to detected losses, both short term and long term.

In this particular embodiment, the video data is coded according to a hierarchical video coding on a set of hierarchical level (in English "Scalable Layer"). The video data may be encoded for example in accordance with a data compression video coding standard such as H.264 or SVC, where the notion of NAL entity described in the introductory part of the present description is involved. The control and decoding method is implemented on the side of a client device in the sense of the client / server architecture. As illustrated in FIG. 5, a video transmission system 15 adapted to implement the present invention comprises a server 11 and a client device 12. The server 11 comprises a storage unit 14 adapted to store, for example, an encoded video. The server 11 further comprises a computing unit 13 adapted to execute programs as well as a network unit 15 adapted to transmit video data on the network. A data acquisition unit (not shown) such as a video camera can be added to capture live video, the computing unit 13 supporting the compression of the video data. The client device 12 likewise comprises a storage unit 16 adapted to store programs of the client device, and in particular a program for decoding and displaying a video. A computing unit 17 is adapted to execute these programs. In particular, the storage unit 16 of the client device comprises one or more programs that include sequences of instructions for implementing the control method and the decoding method according to the invention, when this or these programs are The client device 12 also includes a network unit 18 adapted to receive data packets transmitted over the communication network. Finally, the client device 12 also includes a display unit 19 adapted to display the decompressed video data. Here we place ourselves in a communication network in which the data are transported on the network according to the RTP protocol. Thus, each RTP packet transmitted on the communication network carries a slice (SVC). In particular, each packet includes data on the sequence number of the RTP packet and the time stamp of the RTP packet; these data are placed in the RTP header. Furthermore, the NAL comprises in a header called SVC Extension Header (in English "SCV NAL Extension Header"), the parameters DId, QId and TId defined in the introductory part of the present description. all of this information is managed at the level of the client device 12 to control the loss of data as will be described later .. Reference will now be made to Figure 6 an embodiment of the control and decoding method implemented by the client device of the communication network Here, the server continuously transmits the SVC stream on a set of 25 multicast sessions, each hierarchical level defined by a spatial layer level DId of Qid quality layer and TId time layer is associated with a multicast session.Thus, each multicast session is associated with a triplet Did, Qid, Tid, defined by the vale respectively of the DId, QId, TId parameters contained in the NAL SVC header.

As illustrated in FIG. 6, the method for controlling and decoding a video sequence begins with a subscription step 600 by the client device at a first multicast session that contains the lowest hierarchical level.

This subscription step 600 is followed by an initialization step 601. During this initialization step 601, the size of a Long Term Analysis Window (WS) is determined. . As will be apparent from the rest of the method, this long-term analysis window WS is a sliding window, used to enable the client device to analyze the long-term effects of data loss on decoding to adapt the client device subscription to different multicast sessions. For example, the size of the long-term analysis window WS may be initialized to the time length of two GOPs (acronym for "Group Of Pictures") containing a predetermined number of images, for example 16 or 32) noted in the following 2A. Each RTP packet carries a NAL or a portion of NAL (if the NAL is too big). In SVC, a GOP starts with a NAL corresponding to the lowest spatial hierarchy index (Did = O), the lowest time hierarchy index (Tid = O), and the lowest quality index. (Qid = O). This GOP will end when we receive the next NAL with the same indices (Did = O, Tid = O, Qid = O). These are the data contained in the NAL Header and the SVC Extension Header, which identify the beginning and end of a NAL. During the initialization step 601, an LST variable (acronym for the term "Low Subscription Time") makes it possible to set the duration that elapses between two attempts by the client device to subscribe to a new multicast session. comprising a hierarchical level of video data higher than that of the multicast sessions to which it is already subscribed. The value of this variable LST is for example initialized to a GOP, LST = A.

This initialization step 601 is followed by a reception step 602 corresponding to the reception of a NAL. A loss detection step 603 is then implemented. If losses are detected, the identification of the video data affected by these losses is implemented during an identification step 605. This identification step 605 consists in locating the lost data, that is to say the positions of the NALs lost in the group of GOP images and to store these positions in the memory of the client device. These loss detection and identification steps 603, 605 are detailed with reference to FIGS. 7 and 8. As shown in FIG. 7, the detection / location process of steps 603, 605 of FIG. test 701 which consists in comparing the sequence number, denoted SeqNum, of the last packet received, and the sequence number, denoted SeqNum_, of the packet received immediately before the last packet received. If the difference between these two sequence numbers is 1 (test 701 positive), it can be deduced that no loss has occurred. In this case, the detection / location process ends. Otherwise (test 701 negative), it is already known at this stage that at least one packet is lost, without however yet knowing which NAL is affected.

In this case, a test 703 is performed, comparing the Timestamp, the timestamp, of the last received packet, and the Timestamp_ timestamp, of the received packet immediately before the last received packet. If these two consecutive packets do not have the same value of time stamp (negative test 703), step 801, described later in connection with FIG. 8, is carried out. On the other hand, the two consecutive packets have the same value. time stamp value (test 703 positive), we compare in the next test 707 Did and Did_.

The notation Did represents the value of the parameter DId of the last packet received and the notation Did_ represents the value of the parameter DId of the packet received immediately before the last received packet.

If Did = Did, we test if Qid = Qid_ during the next test 709. The notation Qid represents the value of the parameter QId of the last received packet and the notation Qid_ represents the value of the parameter QId of the packet received immediately before the last received packet. If the test 709 is positive (Qid = Qid_), this means that at least one slice is missing in the same AU, in the same spatial layer and in the same quality layer as those of the NAL received last (event 725 ). If the test 709 is negative, one tests if Qid = Qid_ + 1 during the test 711. If the test 711 is positive (Qid = Qid_-F1), one can deduce that it misses at least a slice in the same AU and in the same spatial layer and with a quality layer equal to Qid or Qid_ (event 727). If the test 711 is negative (Qid is not equal to Qid_-F1), it can be deduced that the NALs for which QId is between Qid_ and Qid are lost (event 713).

If the two consecutive packets have values different from Did (test 707 negative), one tests during a test 715 if Did = Did_-F1. If this is not the case (test 715 negative), we can deduce that the NALs for which Did is between Did and Did_ are lost (event 721). Otherwise, that is to say if Did = Did_ + 1 (positive test 715), we test if Qid = 0 (test 717). If Qid is not zero (test 717 positive), it means that NALs with QId = 0 and QId <Qid have been lost (event 723). Otherwise (ie if Qid = 0, ie test 717 negative), we can deduce that we are in the same AU, in a new spatial layer and with QId = O. We thus deduce that at least one slice with QId = 0 is missing (event 719).

As shown in FIG. 8, the complementary part of the detection / location process starts when the two consecutive packets do not have the same time stamp value (negative test 703). A test 801 is then performed to determine whether the difference between the two time stamp values (denoted Timestamp and Timestamp_) is equal to a unit of time, denoted by A, which corresponds to the time length of a group of GOP images. .

If this is not the case (test 801 negative), we can consider that all the NALs whose timestamp values are between Timestamp_ and Timestamp, which should have been received between Timestamp_ and Timestamp, have been lost ( event 803).

The test 801 and the step 803 are followed by the test 805, which consists in determining if Did <Did_. If Did <Did_ (test 805 positive), one determines during a test 807 if Did_ and Qid_ take their maximum value, that is to say one tests if Did_ = MaxDid and Qid_ = MaxQid. If this is not the case (negative 807 test), it means that NALs with QId> Qid_ and DId> Did_ are missing at the end of the last AU (event 809). The test 807 and the event 809 are followed by an 811 test which consists in determining if Did takes its minimum value for the current value of Tid, that is to say if Did = minDid (Tid).

If this is the case (test 811 positive), it can be deduced (event 813) that a new AU has been received and that at least one slice is missing, ie with DId = minDid (Tid) in the current AU , either with DId = MaxDid or QId = MaxQid in the AU received previously. If the 811 test is negative, ie if Did is different from minDid (Tid), we can deduce that it lacks spatial layers with DId <Did at the beginning of the current AU (event 815) . If the test 817 is positive, that is to say if Did = Did_ = 0, it is determined during a test 819 if the maximum value of DId is zero (ie MaxDid = 0). If this is the case, proceed to test 821 described below. Otherwise, it means that only the NALs corresponding to DId = 0 have been received for the previous AU, and that the NALs of this AU having higher DId values are lost (event 820). The test 821 is to examine if Qid <Qid_. If this is the case (test 821 positive), it is determined during a test 823 if Qid_ = MaxQid. If this is the case (positive test 823), it can be deduced (event 833) that the first slices with QId Qid are missing in the current AU.

If the 823 test is negative, it is followed by a test 825 where we examine if Qid = minQid (Tid). If this is not the case (negative test 825), it can be deduced that the first quality layers of the current spatial layer in the current AU are missing (event 827). If, on the other hand, Qid = minQid (Tid) (test 825 positive), this means that the last quality layers are missing in the previous AU (event 835). If we do not have Qid <Qid_ (test 821 negative), we examine during a test 829 if Qid = Qid_ = O. If this is the case, it means that the first slices of the current AU with QId = 0 are missing (event 831). If not (829 negative test), we can deduce (event 837) that the first quality layers of the current AU are missing. If the test 817 is negative, that is to say if Did and Did_ are not equal and are non-zero, we return to the test 807 described above. At the end of this process, the NALs affected by data loss are thus perfectly identified and localized. Returning to FIG. 6, a test 607 then checks whether the last received NAL corresponds to a new group of GOP images or whether it still belongs to a previously detected group of images. If the group of GOP images is not finished, it returns to the receiving step 602 to receive the other NALs of the same GOP. On the other hand, if a new GOP image group is detected, it is checked in the test step 608 whether errors have occurred on the previous GOP image group. If not, the test step 608 is followed by a subscribe step 609 in which the client device can subscribe to an additional multicast session if it is not already subscribed to all of the subscriptions. multicast sessions available on the communication network. In practice, during the subscription step 609, it is checked whether or not the duration elapsed since the last subscription attempt by the client device is greater than the variable LST.

If this duration exceeds the current value of LST, then a subscription attempt by the client device can be implemented. On the other hand, if the elapsed time is less than the current value of LST, no subscription attempt is implemented on the communication network. Moreover, a decoding step 610 of the video data received in the GOP image group is implemented. The subscription step 609 and the decoding step 610 are followed by the initialization step 601 previously described and all of the preceding steps are reiterated. Otherwise, if at the end of the test step 608, errors have been detected in the preceding GOP image group, a step of selecting a decoding technique according to the location of the video data affected by data loss is implemented. In this embodiment, it is a question of selecting the hierarchical levels that can be decoded taking into account the impact of the detected and localized data losses on a first set of corresponding images, in this embodiment and in a manner that non-limiting, to a group of GOP images. This selection step 611 of the hierarchical levels to be decoded is described in detail with reference to FIGS. 9 and 10. The hierarchical level (s) selection process (s) 611 consists first of all in initializing the variables Did, Qid, Tid and NumFrame at value 0 (step 905). The NumFrame variable indicates the number of the processed image. Then, during a test 907, the NAL which has been identified as lost during the execution of the error detection / localization process is searched. If the NAL having a given value of Did, Qid and Tid (which we note NAL (Did, Qid, Tid)) is lost, the corresponding image is marked as lost (even if only a slice of this image is lost ) in a step 913. This step is followed by a test 915 which will be described below. If the test 907 is negative, that is to say that the NAL considered NAL (Did, Qid, Tid) is not lost, it is determined during a test 909 if the reference image of this NAL has been lost. If this is not the case (test 909 negative), the test 915 described below is carried out. If, on the other hand, the reference image of this NAL has been lost (test 909 positive), the image concerned is marked as lost (step 911).

The 915 test consists of determining whether all the quality layers have been analyzed. For this, we test if Qid MaxQid, where MaxQid designates the highest quality layer. If all the quality layers have not yet been analyzed, ie if Qid MaxQid (test 915 positive), we increase Qid, for example by incrementing by one unit the value of Qid (step 916) then the process continues with a new NAL (return to test 907). If all the values of Qid have been tested (test 915 negative), we pass test 917, in which we test if all the spatial layers have been analyzed. For this, we test if Did MaxDid, where MaxDid designates the highest spatial layer. If all the spatial layers have not been analyzed (test 917 positive), during a step 919 we increase Did, by incrementing for example by one unit the value of Did, and we attribute to Qid the minimum authorized value of Qid for the current value of Tid (noted minQid (Tid)). If, on the other hand, all the values of Did and Qid for the current AU have been analyzed, a test 921 examines whether all the images in the GOP for the current temporal layer (denoted Tid) have been analyzed. For that, one tests if NumFrame NumFrame (Did, Tid). If the GOP images for Tid have not all been analyzed (positive test 921), the process continues with the next image and the NumFrame variable is incremented by one (step 923). In addition, in step 923, the parameter DId is set to its minimum authorized value for the current value of Tid (this minimum value is denoted MinDid (Tid)). In addition, in step 923, the QId parameter is set to its minimum allowed value for the current value of Tid (this minimum value is noted MinQid (Tid)).

When all the images of the current temporal layer have been analyzed (negative test 921), a test 925 examines whether all the temporal layers have been tested. For this, we test if Tid MaxTid, where MaxTid designates the highest time layer. If all the temporal layers have not been analyzed (test 925 positive), go to step 929, where we go to the upper temporal layer (Tid + 1) and where the parameters DId and QId are fixed to their values. respective minimum values for the current value of Tid (these minimum values are respectively denoted minDid (Tid) and minQid (Tid).

When all the temporal layers have been analyzed (negative test 925), all the lost images, either directly (by packet loss) or indirectly (by error propagation) are identified (state 931). The process of selecting the layers to be decoded can then begin. This process is illustrated on the flowchart in Figure 10.

The selection process starts at a step 1001 by the initialization of Did, Qid and Tid at the value O. Moreover, the process starts with the first image having the current values of Did and Tid (ie NumFrame (Did, tid) = 0). These variables will allow to analyze (in English "to parse") the whole structure of the GOP. In addition, a variable Nloss (x, y, z) designating the number of losses is also initialized to the zero value during step 1001 (the variables x, y and z taking respectively all the possible values of Did, Qid and tid). Then during a test 1003, we examine if the image considered is lost. If it is not lost (test 1003 negative), the image concerned is marked as displayed in the representation of the structure of the GOP (step 1005). Step 1005 is followed by a test 1017 described later. If, on the other hand, the image considered is lost (test 1003 positive), the variable Nloss (Did, Qid, Tid) of one unit is incremented (step 1007). Then during a test 1009, we examine if the value of the Nloss variable (Did, Qid, Tid) is strictly greater than a threshold which depends on the values of Did, Qid and Tid.

This threshold is noted Th (Did, Qid, Tid). It represents the acceptable amount of losses for a given triplet of values of Did, Qid and Tid. Indeed, each triplet represents a hierarchical level. Since some levels are higher than others, the acceptable level of loss can not be the same for all hierarchical levels. Thus, the threshold value Th (Did, Qid, Tid) has a value dependent on the hierarchical level associated with the triplet of values Did, Qid, Tid. If the threshold is not exceeded (test 1009 negative), we go to test 1017. Otherwise (test 1009 positive), all the images with DId Did, QId Qid and TId Tid are marked as not displayed during a step 1011 Then, during a step 1013, the values of MaxQid, MaxTid and MaxDid are modified. We avoid here the analysis (in English "parsing") of the entire structure of the GOP if we already know that the quality of the following layers will not be acceptable.

Thus, in step 1013, the value of MaxQid for the values of DId greater than or equal to Did and for the values of TId greater than or equal to Tid is fixed either at the value (Qid ù 1) or at the null value if Qid = 0. Also in step 1013, the value of MaxTid for the values of DId greater than or equal to Did is fixed either at the value (Tid ù 1) or at the value zero if Tid = 0 and the value of MaxDid is set either to the value (Did ù 1) or to the null value if Did = O. Step 1013 is followed by a test 1019 which will be described later. In test 1017, we examine whether all the images with the current triplet (Did, Qid, Tid) were processed. If the images corresponding to this triplet have not yet been processed (test 1017 negative), we go to the next image (step 1015) and then return to the test 1003 described above. When all the images having the current triplet (Did, Qid, Tid) have been processed (test 1017 positive), one passes to the test 1019. During the test 1019, one examines if Did reached its maximum value.

For that, one tests if the Did value is lower or equal to MaxDid.

If Did has not reached its maximum value (test 1019 positive), increase Did, for example by incrementing the value of Did (step 1021) by one unit, then return to test 1003. If Did has reached its value MaxDid maximum (test 1019 negative), we perform a test 1023, consisting of examining whether Tid has reached its maximum value for the current value of Did. For that, one tests if the value Tid is lower or equal to MaxTid (Did). If Tid has not reached this maximum value (test 1023 positive), Tid is increased, for example by incrementing by one unit the value of Tid (step 1025), then returns to the test 1003. Moreover, at step 1025, Did is reset to its minimum value for the given value of Tid. If Tid has reached its maximum value for the current value of Did, that is Tid = MaxTid (Did) (test 1023 negative), a test 1027 is carried out, consisting in examining whether Qid has reached its maximum value for the current values of Did and tid. For that, one tests if the value Qid is lower or equal to MaxQid (Did, Tid). If Qid has not reached this maximum value (test 1027 positive), the value of Qid is increased, for example by incrementing the value of Qid by a unit (step 1029), then returns to the test 1003. Moreover, in step 1029, Did is reset to its minimum value for Tid = 0 and Tid is reset to zero. If Qid has reached its maximum value for the current values of Did and Tid, ie Qid = MaxQid (Did, Tid) (test 1027 negative), the process of selecting hierarchical levels 611 is finished.

It should be noted that, in the flowcharts of FIGS. 7 to 10, the descriptions of the tests consisting in testing, whether a value is reached and / or exceeded, or whether a value is strictly greater than a threshold, or whether all the images have been processed, etc., have been given as non-limiting examples, in accordance with the particular embodiments described. As a variant, these tests could equally well consist in testing the inverse condition, namely, respectively, or if the value considered is not reached and / or is not exceeded, or if the value considered is less than or equal to threshold, or if there is at least one image not yet processed, etc. The processes illustrated in Fig. 10 are followed by the NAL decoding step 613 illustrated in Fig. 6.

Before decoding the current NAL, the client device verifies whether the image concerned has been marked as lost, in the structure of the GOP stored in the storage unit 6 of the client. If the current image has been marked as lost, the SVC decoder rejects the current NAL without decoding it. If this image is not marked as lost, the NAL is decoded. The decoding step 613 is then followed by the test step 615 on the size of the sliding analysis window WS. Thus, in test step 615, it is checked whether the size of the sliding analysis window WS is reached, that is, whether a sufficient quantity of GOP image groups has been analyzed at steps 603 and 605. If not, all the steps are repeated from step 602 while receiving additional NALs. On the other hand, if the size of the sliding analysis window WS is reached, that is to say if WS is equal here to 2A, a loss evaluation and adaptation step of the subscription 617 is set 20 from the examination of the impact of long-term losses on GOP image groups. This loss evaluation and subscription adaptation step is described in detail with reference to FIG. 11. In principle, the evaluation of the importance of the data losses on the video data received over the long term. , that is to say on a second set of images of length, here 2A, greater than the length, here A, of the first set of images, reuses the principle described above with reference to FIG. FIG. 9 makes it possible to evaluate the losses on a first set of small images, corresponding here to a group of GOP images, the loss evaluation step as described later with reference to FIG. to evaluate the losses on a second set of images of greater length, corresponding to the size of the sliding analysis window WS, here equal to two groups of GOP images. It will thus be noted that this second set of images, corresponding to the sliding analysis window WS, consists of a set of successive GOP image groups including the first set of GOP images analyzed above, on which losses of data was detected. This loss evaluation and adaptation step of the subscription 617 begins with an initialization step 1100. The client device initializes the values Did, Tid and Qid to their maximum values, these values corresponding to the highest hierarchical level. received in the highest multicast session. Did = MaxDid; Qid = MaxQid (Did) and Tid = MaxTid (Did). All images corresponding to these values of Did, Qid and Tid in the set of GOP image groups corresponding to the sliding analysis window WS will have to be analyzed. A variable n is used to enumerate all the images in the highest hierarchical level, this variable n being initialized to zero. On the other hand, a variable N is initialized to the value NumFrame (Did, Tid, Qid) corresponding to the number of images in the highest hierarchical level. Finally, the Nloss value is initialized to zero and represents the number of images lost in the highest level. In a first step 1101, a first image 25 is considered in the given hierarchical level. A verification step 1103 makes it possible to check whether the image is lost. If not, this verification step 1103 is followed by a step 1108 in which it is checked whether all the images have been analyzed. In practice, it is checked whether n <N. If all the images have been verified (i.e. if the test step 1108 is negative), the test step 1108 is followed by a step d subscription adaptation 1113 of maintaining the current subscription level of the client device. On the other hand, if certain images have not been analyzed, at the end of the test step 1108, an incrementation step 1109 is implemented in order to increase by one unit the index n and it is repeated step 1103 from a next image. If at the end of the test step 1103, the image is detected as lost, the number of Nloss losses is increased by one unit in the incrementation step 1105.

A comparison step 1107 is used to compare the number of losses Nloss with the threshold value Th (Did, Qid, Tid), described above, corresponding here to the acceptable number of losses in the highest hierarchical level. If the number of Nloss losses is greater than the threshold value, the subscription adaptation step 1111 is implemented so that the client device makes the decision to unsubscribe from the highest hierarchical level. Otherwise, the comparison step 1107 continues with the steps 1108 and following described previously.

Thus, returning to FIG. 6, a test step 619 makes it possible to identify the decision that was taken during the subscription adaptation steps 1111, 1113 in FIG. 11. Thus, if the unsubscription at a hierarchical level given, the test step 619 is followed by the test step 625 where it is checked whether the size of the sliding analysis window WS is different from the value A corresponding to the time length of a group d GOP images. If so, the size of the sliding analysis window WS is reduced by a value A such that WS = WS ùA at a subtraction step 627. On the other hand, if at the end of the comparison step 625 , the size of the sliding analysis window WS is already equal to the minimum value A, it retains its value.

In contrast, the LST variable corresponding to the period of time between two attempts to subscribe to a higher-level multicast session by the client device is increased by the A value, which is the size of a GOP at an addition step 629 such that LST = LST + A. The whole process is then reiterated from the step of receiving a NAL 602. On the other hand, if at the end of the test step 619, the adaptation of the subscription consists in maintaining the same subscription level, the test step 619 is followed by an addition step 621 in which the size of the sliding analysis window WS is increased by an A value corresponding to the size of a group of GOP images. The addition step 621 is followed by a reset step 623 of resetting the value LST to A. The entire process is then reiterated from the receiving step of a NAL 602. Thus, the length of the sliding analysis window WS varies according to the adaptation of the subscription by the client device to the subset of multicast sessions. The length of the sliding analysis window WS is decreased, if it has not reached its minimum size A, when the client device unsubscribes from at least one multicast session. In contrast, the length of the WS rolling scan window is increased when the client device maintains its subscription to a subset of multicast sessions. In this embodiment, the length of this sliding analysis window WS is increased or decreased in each case by the length A of a group of GOP images. On the other hand, the LST time interval between two subscription attempts by the client device to an additional multicast session is increased when the length of the sliding analysis window WS reaches its minimum prefixed value, here equal to A corresponding to a GOP.

FIGS. 12 and 13 illustrate the principle obtained by such a control method in the subscription of a client device to a subset of multicast sessions. As illustrated in FIG. 12, during the first three successive periods of time T1, T2, T3, of length A = GOP, the client device subscribes to additional multicast sessions corresponding to hierarchical levels that are higher and higher ( level 0, level 1, level 2). Indeed, since the bit rate linked to the transmission of these multicast sessions remains lower than the available bandwidth or bandwidth on the communication network, there are no errors and losses of detected data (which corresponds to a negative response to the error presence test step 608). Then, when the client device seeks to subscribe to an additional multicast session (corresponding to the hierarchical level 3), errors appear in the data transmission since the accumulated bitrate exceeds the available bandwidth on the communication network. However, as long as the amount of errors evaluated in the highest multicast session is considered acceptable, i.e. the number of evaluated data losses does not exceed the predetermined threshold value which depends on it. given hierarchical level, the control and decoding method keeps unchanged the subscription of the client device, and in particular keeps the subscription to the highest multicast session. On the other hand, as illustrated in FIG. 13, if the number of data losses evaluated at the fourth period T4 is not acceptable for the decoding, the client device continues to receive during this period T4 the four multicast sessions (level hierarchical 0 at hierarchical level 3), although the content of hierarchical level 3 of the highest multicast session is not decoded.

After the fifth period T5, the amount of losses evaluated on the sliding analysis window WS, corresponding to the fourth and the fifth period T4, T5 is considered unacceptable if the number of evaluated data losses exceeds the value predetermined threshold associated with this hierarchical level (here level 3). In this case, the client device unsubscribes from the fourth multicast session (and no longer receives level 3). This unsubscription is visible on the sixth period T6 in Figure 13.

In this case, as explained with reference to FIG. 6, the length of the sliding analysis window WS is reduced by an value A and the client device again attempts to subscribe to the fourth multicast session at the same time. expiry of the sixth period T6. This new subscription causes the seventh period T7 again to lose data whose number exceeds the predetermined threshold value. In this case, the client device unsubscribes again for the eighth period T8 of the fourth multicast session. Since the size of the sliding analysis window WS is equal to its minimum A, the length of the time interval between two LST subscription attempts is increased by an A value so that a new attempt of subscription to the fourth multicast session only occurs in the tenth T10 period. Thus, the control and decoding method allows the client device to space out subscription attempts to a higher level multicast session when this subscription systematically causes congestion in the communication network. On the other hand, through the use of a sliding scan window WS larger than the length of a GOP, it is possible not to systematically react by unsubscribing the client device to a multicast session of higher level when network congestion is short-lived.

A practical embodiment of the invention has previously been described in which the evaluation of the loss of data on the highest hierarchical level received results in the unsubscription of the multicast session transmitting this highest hierarchical level.

However, when significant network congestion occurs, it may be beneficial to allow simultaneous unsubscribing to multiple multicast sessions. One solution is to use the insertion by the server of a priority level associated with each packet of video data transferred.

The priority level inserted in each video data packet depends on the hierarchical level of the inserted video data as well as the dependence of that video data with other video data inserted into different packets. In practice, this priority level can be inserted in the NAL header, in the priority_ID field or in the header of a data packet. In practice, a high priority level will be associated with a multicast session carrying a low hierarchical level. On the other hand, a low priority level will be associated with a multicast session carrying a high hierarchical level.

Conventionally, in a communication network implementing priority levels associated with RTP packets, a router systematically rejects the packets in order of priority when congestion of the communication network occurs. Also, in case of congestion of the communication network, the video data packets having the lowest priority levels will be rejected first. Thus, if a data packet associated with a high priority level is lost, the client device can deduce on the one hand that significant congestion of the network occurs and on the other hand that the video data packets associated with a level lower priority will also be lost. As soon as at the end of the steps of identification and location of losses as described with reference to Figures 7 and 8, it is possible to know exactly what NAL is lost, it is also possible to know the level of priority of each lost NAL, stored for example in the NAL header, in the priority_ID field. Each priority level being respectively associated with a data packet, the multicast session associated with a lost data packet can be identified. It is then possible to tailor the client device subscription to the subset of multicast sessions based on the identified priority level of a video data packet affected by data loss. In practice, subscription adaptation step 617 is adapted to apply sequentially to all hierarchical levels affected by losses, starting with the lowest hierarchical level. If the number of evaluated data losses exceeds the predetermined threshold value of loss of data associated with the analyzed hierarchical level, when the subscription is adapted, not only is the multicast session corresponding to the analyzed hierarchical level rejected, but also the set of multicast sessions transmitting video data hierarchical levels above the hierarchical level 20 analyzed. Thus, the client device unsubscribes to several multicast sessions at a time.

Claims (6)

REVENDICATIONS1. A method of control by a client device (12) of the transfer of a video sequence transmitted over a set of multicast sessions by a server (11) in a communication network, said video sequence being coded according to a hierarchical video coding on a set of hierarchical levels, each hierarchical level being transmitted over the communication network respectively by a multicast session, said client device (12) subscribing to at least a subset of multicast sessions, characterized in that it comprises the following steps: - detection (603, 605), on a first set of images of said video sequence, of data losses affecting video data of said first set of images; selecting (611) a decoding technique according to the location of the video data affected by said data loss in said first set of images; detecting (617), on a second set of images (WS) of said video sequence, the length of said second set of images being greater than the length of said first set of images, data loss affecting video data said second set of images; - evaluating (617) the importance of the loss of data affecting the video data of said second set of images (WS); and - matching (617) the subscription by said client device (12) to the subset of multicast sessions as a function of the amount of data loss affecting the video data of said second set of images (WS) . 2. Control method according to claim 1, characterized in that said step of selecting (611) a decoding technique is adapted to select at least one hierarchical level not to be decoded. 30 3. Control method according to one of claims 1 or 2, characterized in that in said evaluation step (617), the importance of the loss of data is evaluated on video data of a given hierarchical level and in that at said adaptation step (617), the client device (12) unsubscribes from a multicast session transmitting said given hierarchy level when the number of evaluated data losses exceeds a predetermined threshold value Th ( Did, Qid, Tid) data loss. 4. Control method according to claim 3, characterized in that said predetermined threshold value Th (Did, Qid, Tid) has a value dependent on said given hierarchical level. 5. Control method according to one of claims 3 or 4, characterized in that said adaptation step (617), said client device (12) also unsubscribes from at least one multicast session transmitting respectively hierarchical video data higher than said given hierarchical level. 6. The checking method as claimed in claim 1, wherein said server inserts priority levels respectively into packets of video data respectively transferred over a set of multicast sessions, the priority level inserted in a packet of packets. video data being dependent on the hierarchical level of said video data of said packet, characterized in that the control method comprises the following steps: - identification of the priority level associated with a video data packet affected by said loss of data detected during said step detecting (617) a second set of images (WS) of said video sequence; and - adapting the subscription by said client device (12) to the subset of multicast sessions according to said identified priority level. 9. Control method according to one of claims 1 to 6, characterized in that it comprises a step of varying (621, 627) the length of said second set of images (WS) according to the adaptation the client device (12) subscribing to the multicast session subset at said matching step (617). A control method according to claim 7, characterized in that at said step of varying (627), the length of said second set of images (WS) is decreased when at said matching step (617), said client device (12) unsubscribes from at least one multicast session. 9. Control method according to one of claims 7 or 8, characterized in that at said step of variation (621), the length of said second set of images (WS) is increased when at said step of adaptation (617), said client device (12) maintains its subscription to said subset of multicast sessions. 10. Control method according to one of claims 7 to 9, characterized in that the first set of images is a group of images (GOP), the second set of images (WS) consisting of a set of successive image groups including said first set of images. 11. Control method according to claim 10, characterized in that the length of said second set of images (WS) is increased or decreased by the length (A) of a group of images (GOP). 12. Control method according to one of claims 1 to 11, characterized in that, when the length of said second subset of images (WS) is equal to a minimum value prefixed (A), a time interval (LST) between two subscription attempts by said client device (12) to an additional multicast session, in addition to said subset of multicast sessions, is increased. 13. A method of decoding by a client device (12) a video sequence transmitted over a set of multicast sessions by a server (11) in a communications network, said video sequence being encoded according to a hierarchical video coding on a set of hierarchical levels, each hierarchical level being transmitted over the communication network respectively by a multicast session, said client device (12) subscribing to at least a subset of multicast sessions, characterized in that it comprises the following steps: - detection (603, 605), on a first set of images of said video sequence, of data losses affecting video data of said first set of images; decoding (613) a selection of hierarchical levels according to the location of the video data affected by said data losses in said first set of images; detecting (617), on a second set of images (WS) of said video sequence, the length of said second set of images being greater than the length of said first set of images, data losses affecting video data of said second set of images; - evaluating (617) the importance of the loss of data affecting the video data of said second set of images; and - adapting (617) the subscription by said client device (12) to the subset of multicast sessions as a function of the amount of data loss affecting the video data of said second set of images (WS). 14. Device for controlling the transfer of a video sequence transmitted on a set of multicast sessions by a server (11) in a communications network, said video sequence being coded according to a hierarchical video coding on a set of hierarchical levels, each hierarchical level being transmitted over the communication network respectively by a multicast session, said client device (12) subscribing to at least a subset of multicast sessions, characterized in that it comprises: detection means (16, 17, 18), on a first set of images of said video sequence, of data losses affecting video data of said first set of images; selection means (16, 17, 18) of a decoding technique as a function of the location of the video data affected by said data losses in said first set of images; detection means (16, 17, 18) on a second set of images (WS) of said video sequence, the length of said second set of images being greater than the length of said first set of images, losses data affecting video data of said second set of images; evaluation means (16, 17, 18) of the importance of the data losses affecting the video data of said second set of images (WS); and means for adapting (16, 17, 18) the subscription by said client device (12) to the subset of multicast sessions as a function of the importance of the data losses affecting the video data of said second set of images. 15. Control device according to claim 14, characterized in that it comprises means for varying (16, 17, 18) the length of said second set of images (WS) according to the adaptation of the subscription to the subset of multicast sessions. 16. Control device according to one of claims 14 or 15, said server (11) inserting priority levels respectively in video data packets respectively transferred on a set of multicast sessions, the priority level inserted in a video data packet being dependent on the hierarchical level of said video data of said packet, characterized in that it further comprises: - identification means (16, 17, 18) of the priority level associated with a video data packet affected by said loss of data detected during said detecting step on a second set of images (WS) of said video sequence; and means for adapting (16, 17, 18) the subscription by said client device (12) to the subset of multicast sessions as a function of said identified priority level. 19. Client device in a communication network, comprising calculation means (17), storage means (16) and communication means (18) with said network, said calculation means (17) cooperating with said means of storage (16) and said communication means (18) with the network for carrying out the control method according to one of claims 1 to 12 and / or the decoding method according to claim 13. 20. Means for storage of information readable by a computer or a microprocessor retaining instructions from a computer program, characterized in that it makes it possible to implement a control method according to one of claims 1 to 12 and / or a method decoding device according to claim 13. 19. Computer program product that can be loaded into a programmable device, characterized in that it comprises instruction sequences for implementing a control method according to one of claims 1 to 12 and / or a decoding method according to claim 13, when this program is loaded into and executed by the programmable apparatus.