FR3020168A1 - ROTATING WING DRONE WITH VIDEO CAMERA DELIVERING STABILIZED IMAGE SEQUENCES - Google Patents

Abstract

The drone (10) includes a camera with a fisheye type hemispherical field lens pointing in a fixed direction (A) relative to the body of the drone. A reduced-size capture zone (36) is extracted from the image formed by this objective (42), the position of this zone being a function of signals delivered by an inertial unit measuring the Euler angles characterizing the attitude of the drone compared to an absolute landmark. The position of this zone is dynamically modified in a direction (44) opposite to that of the changes of attitude (38) of the drone detected by the inertial unit. The raw pixel data is then processed to compensate for the geometric distortions introduced by the fisheye lens onto the image collected in the region of the capture area.

Description

The invention relates to rotary wing drones such as quadcopters and the like. These drones are provided with multiple rotors driven by respective engines controllable in a differentiated manner to control the drone attitude and speed. A typical example of such a drone is AR.Drone 2.0 from Parrot SA, Paris, France, which is a quadricopter equipped with a series of sensors (accelerometers, three-axis gyrometers, altimeter), a front camera capturing a image of the scene towards which the drone is directed, and a vertical aiming camera capturing an image of the terrain overflown. The WO 2010/061099 A2 and EP 2 364 757 A1 (Parrot SA) describe such a drone and its driving principle via a phone or multimedia player with touch screen and built-in accelerometer, for example a cellular phone. iPhone type or iPad type multimedia tablet (trademarks of Apple Inc., USA). These devices incorporate the various control devices necessary for the detection of control commands and the bidirectional exchange of data with the drone via a Wi-Fi (IEEE 802.11) or Bluetooth (registered trademarks) wireless network link. They are further provided with a touch screen displaying the image captured by the front camera of the drone, with a number of symbols superimposed enabling the activation of commands by simply touching the finger of the user on this touch screen. The front video camera of the drone can be used for "immersive mode" control, that is to say where the user uses the image of the camera in the same way as if he were himself even on board the drone, it can also be used to capture sequences of images of a scene to which the drone is moving, allowing the user to use the drone in the same way as a camera or a camcorder that, instead of being held by hand, would be carried by the drone.The collected images can be recorded and broadcast, posted on web hosting video sequences, sent to other users, shared on social networks, etc. These images are intended to be recorded and communicated, it is desirable that they have the least possible defects, including defects resulting from parasitic movements of the drone, which will cause oscillations and unwanted jumps of the image capt By the camera in particular, with the camera pointing in the main direction of the drone, any movement around the pitch axis (or yaw axis), which is perpendicular to the axis of the camera, will produce on the image vertical oscillations (respectively, horizontal) strongly degrading the readability and the quality of the captured image. Likewise, any movement around the roll axis (the axis of the camera) will cause the image to rotate in one direction or the other, affecting its readability. However, the movements of a rotary wing drone such as a quadricopter, whether user controlled or servo-controlled by an autopilot, result mainly from rocking movements around the pitch axes (forward / backward movements). and roll (left / right movements), which are inherent in the very principle of operation of such a drone. Specifically, if the drone is commanded to tilt or "dive" downward (tilting at a pitch angle), it will progress forward, with a speed as high as the tilt will be important. Conversely, if it is controlled so as to "pitch up" in the opposite direction, its speed will gradually slow down and then reverse by going backwards. In the same way, for an inclination control along a roll axis the drone will lean to the right or left, causing a linear displacement in horizontal translation to the right or to the left. Any linear movement of the drone forwards or backwards or on the side involves a tilting of the drone, and therefore a corresponding effect of offset, rotation, oscillation ... of the image collected by the camera.

These disturbances may be acceptable in an "immersive pilots" configuration as they are part of the "user experience." On the other hand, if it is a matter of using the drone as a mobile video camera to capture sequences that will be recorded and retrieved later, these parasitic movements are extremely disturbing, with the image a misaligned and unstable horizon, rising and falling in the image as the accelerations and slowdowns of the drone As well as parasitic rotations and other various artifacts, EP 2 613 214 A1 (Parrot) discloses a method of piloting a drone for shooting in a user-selected mode such as front or side tracking, panoramic or crane plane, defining a trajectory to print to the drone.When the drone stabilized on the prescribed trajectory, the video shooting is activated and the trajectory is stabilized by a open-loop control that avoids the oscillations inherent in servo feedback looping. However, it is in this case to stabilize a trajectory by avoiding unwanted oscillations around a set point by modifying the operation of the attitude control loops of the drone when the movement imposed on it is a translational movement. uniform rectilinear or uniform rotational motion. It is not a question of compensating the displacements of the image resulting from the movements of rocking of the drone during phases of acceleration or deceleration during displacements forward / backward and / or left / right. Various solutions have been proposed to compensate for such displacements in the image. A mechanical solution consists of mounting the camera in a cradle connected to the body of the drone by a suspension with the motorized gimbal and controlled so as to compensate for the rocking movements of the drone. This solution has several advantages, notably to stabilize the image upstream of its capture, and to allow a large amplitude of angle compensation. On the other hand, it involves a complex and heavy mechanical system (which is particularly disadvantageous for a flying object), and the efficiency of the compensation is limited by the maximum acceleration and the speed of the servo motors used.

Another technique, called OIS (Optical Image Stabilization) consists of moving in real time optical elements of the lens of the camera, or the sensor in the focal plane. The stabilization is, here again, operated upstream of the shooting, and this system involves only a very small footprint. On the other hand, the optical design is complex, and the maximum amplitude of angle compensation is limited to a few degrees, with, moreover, a response time sufficient to compensate for the effects of freehand shooting, but too long to compensate for the very sudden movements of a moving drone. Finally, the so-called Electronic Image Stabilization (EIS) technique consists in acquiring on the sensor a fixed zone of greater extent than the capture zone that will be used. Compensation is effected by a translation of the capture zone on the acquisition zone, in the opposite direction of the movement to be compensated, the sensor transmitting only a subpart corresponding to the stabilized image. The implementation of such a compensation is simple. On the other hand, the compensation amplitude is limited by the ratio between the size of the capture zone and that of the acquisition zone, that is to say the effective size of the sensor used. Specifically, the maximum amplitude of angle compensation is limited to a few degrees. In addition, the simple translation of the capture zone is not mathematically sufficient to compensate for a rotation of the camera, because it is not a true correction of the change of perspective induced by rotations. Finally, it is a compensation by post-processing of the image data acquired by the sensor, which does not make it possible to compensate for certain effects such as motion blur and wobble (image waviness). , of low amplitude and high frequency, caused by the vibrations of the drone's engines). The object of the invention is to propose a new technique of image capture by the camera of a drone, in particular a quadrocopter type, which overcomes the aforementioned drawbacks, and provides the following advantages: - large amplitude of angular compensation; - ability to compensate for very fast movements and strong acceleration; - No increase in the size or weight of the embedded elements; - great simplicity of implementation; - compensation for all optical phenomena, including those involving a change of perspective induced by the rotations of the drone; - Compensation of high amplitude and low frequency ripple effects (jelly) and low amplitude and high frequency (wobble).

The invention proposes for this purpose a system applicable to a rotary wing drone of known type, for example according to the aforementioned EP 2,613,214 A1, comprising a camera, provided with a lens pointing in a fixed direction relative to the body of the drone as well as a digital sensor collecting the image formed by the objective, and an inertial unit capable of measuring the Euler angles characterizing the attitude of the drone with respect to an absolute terrestrial reference. In a characteristic manner of the invention, the objective is a fisheye type hemispherical field objective, and the drone further comprises: extracting means, receiving as input a selection signal defining the position of a zone of reduced size capture over the range of the sensor, and outputting raw pixel data corresponding to the capture area; servo-control means, receiving as input at least one Euler angle delivered by the inertial unit, and capable of modifying the selection signal dynamically in a direction opposite to that of a change of attitude of the drone detected. by the inertial unit and characterized by a corresponding variation of said at least one Euler angle; image processing means, receiving as input the raw pixel data delivered by the extractor means and outputting corrected pixel data, compensated for the geometric distortions introduced by the fisheye lens onto the image collected by the sensor; in the region of the capture zone selected by the extracting means; and output means for outputting the corrected pixel data for transmission to a display or video recording device. According to various advantageous subsidiary characteristics: the servo-control means are able to modify the selection signal so that the pixel data delivered by the extracting means correspond to an image centered on the horizon, or centered on a fixed orientation relative to the horizon; the camera is mounted in the drone so that the scanning direction of the digital sensor is oriented parallel to the pitch axis of the drone; said at least one Euler angle is a pitch angle of the drone, and the servo means are able to modify the selection signal so as to translate the capture zone into a first direction, parallel to an axis. main sensor; said at least one Euler angle is a yaw angle of the drone, and the servo means are able to modify the selection signal so as to translate the capture zone in a second direction, perpendicular to said first direction; the digital sensor is a scanning sensor delivering the pixel data line by line, and the image processing means are able to apply to each line of the sensor an additional correction capable of compensating for the relative displacements of pixels of a line; to the next induced by rotations of the drone around a yaw, pitch and / or roll axis. We will now describe an exemplary embodiment of the drone according to the invention, with reference to the accompanying drawings in which the same references indicate from one figure to the other identical or functionally similar elements. Figure 1 is an overview showing the drone and associated remote control device for remote control.

Figures 2a and 2b illustrate the changes in camera sighting directions caused by a forward tilting of the drone, for example during an acceleration phase. FIG. 3 is a block diagram of the various servo control and steering control elements of the drone, as well as correction of the displacement of the image according to the technique of the invention. Figure 4 is an example of an image formed on the camera sensor of the drone. FIG. 5 illustrates the successive steps of windowing and correction of the distortions corresponding to the treatment according to the invention applied to the image of FIG. 4.

Figure 6 illustrates another example of an image of a scene captured with the drone camera. Figure 7 illustrates various views that can be extracted from the overall image of Figure 6, after windowing and distortion correction.

Figure 8 illustrates the deformations made to the central image illustrated in Figure 7 in case of rolling motion. Figure 9 illustrates how to optimally position the sensor to effectively compensate for changes in attitude of the drone around its pitch axis.

Figure 10 is the counterpart of Figure 9 for rotations around a yaw axis. Figures 11, 12 and 13 illustrate the deformations of wobble and jelly type observable on the image of a checkerboard, and which can be compensated for the correction of distortions according to the teachings of the invention. We will now describe an example of implementation of the invention. In FIG. 1, the reference 10 designates in a general manner a drone, which is for example a quadricopter such as the AR.Drone 2.0 model of Parrot SA, Paris, France, described in particular in WO 2010/061099 A2 and EP 2 364 757 Al supra. The drone 10 comprises four coplanar rotors 12 whose engines are controlled independently by an integrated navigation system and attitude control. It is provided with a first camera 14 with a frontal aiming to obtain an image of the scene towards which the mirror is directed. The drone also includes a second vertical aiming camera (not shown) pointing downwards, capable of capturing successive images of the terrain overflown and used in particular to evaluate the speed of the drone relative to the ground. Inertial sensors (accelerometers and gyrometers) make it possible to measure the angular velocities and attitude angles of the drone with precision, ie the Euler angles (pitch 9, roll 6 and yaw p) describing the inclination of the drone relative to a horizontal plane of a fixed terrestrial reference, it being understood that the two longitudinal and transverse components of the horizontal velocity are intimately related to the inclination along the two respective axes of pitch and roll. An ultrasonic range finder disposed under the drone also provides a measurement of the altitude relative to the ground.

The drone 10 is controlled by a remote remote control device 16 provided with a touch screen 18 displaying the image embedded by the front camera 14, with a number of symbols superimposed enabling the activation of control commands by simple contact of the The apparatus 16 is provided with means for radio linkage with the drone, for example of the Wi-Fi local area network (IEEE 802.11), for bidirectional exchange of data. data from the drone 10 to the apparatus 16, in particular for transmitting the image picked up by the camera 14, and from the apparatus 16 to the drone 10 for sending pilot commands.

The remote control apparatus 16 is also provided with inclination sensors for controlling the attitude of the drone by imparting corresponding inclinations to the apparatus along roll and pitch axes (reference may be made to WO 2010 / 061099 A2 cited above for more details on these aspects of the system). The piloting of the drone 10 consists in making it evolve by: a) rotating around a pitch axis 22, to advance or retract it, b) rotation about a roll axis 24, to shift it towards the right or left; c) rotation about a yaw axis 26, to pivot to the right or to the left the main axis of the drone - and therefore also the pointing direction of the front camera 14; and d) downward translation 28 or upward 30 by changing the speed of the gases so as to respectively reduce or increase the altitude of the drone. When these control commands are applied by the user from the remote control device 16, the commands a) and b) pivoting about the pitch axes 22 and the rolling axis 24 are obtained by inclinations of the apparatus 16 respectively around its longitudinal axis 32 and its transverse axis 34: for example, to advance the drone simply tilt the remote control device 16 forward by leaning it about the axis 32, for to depose it to the right it suffices to tilt the remote control device 16 by leaning it about the axis 34 to the right, and so on. The commands c) and d), as for them, result from actions performed by contact of the user's finger 20 with corresponding specific areas of the touch screen 18. The drone also has an automatic and autonomous system of Hover stabilization, activated in particular as soon as the user removes his finger from the touch screen of the aircraft, or automatically at the end of the take-off phase, or in the event of interruption of the radio link between the aircraft and the drone. The drone then moves to a state of levitation where it will be immobilized and automatically stabilized in this fixed position, without any intervention of the user. Figure 2a schematically illustrates, in profile, the attitude of the drone when it is immobile, in a state of levitation. The field covered by a front camera 14 of conventional type has been schematized at 36, for example a camera covering a field of 54 ° and whose axis of vision 8 is centered on the horizon. If, as shown in Figure 2b, the drone progresses forward with a nonzero horizontal velocity, by design the axis 26 of the drone will be tilted forward by an angle 9 (pitch angle) with respect to the vertical V. This inclination forward, shown schematically by the arrow 38, involves a tilt of the same value, shown schematically by the arrow 40, of the axis 6 of the camera relative to the plane of the horizon HZ. We thus understand that as the drone evolves, its accelerations, slowdowns ..., the axis oscillates permanently around the direction of the horizon HZ, which will be reflected in the image captured by movements of permanent oscillation up and down. Similarly, if the drone shifts to the right or to the left, this movement will be accompanied by a pivot about the roll axis 24, which will be reflected in the image by one-way rotations. or in the other of the scene captured by the camera. To overcome this drawback, the invention proposes, instead of using a camera provided with a conventional lens, to provide this camera with a hemispherical field objective of "fisheye" type covering a field of about 180 °. as schematized at 42 in Figure 2a. The image captured by the camera provided with this fisheye lens will certainly undergo the same movements of oscillation and rotation as a conventional camera but, in a characteristic manner of the invention, only part of the field captured by this camera by selecting a particular window, called "capture window", corresponding to the angular sector 36 captured by a conventional camera. This capture area will be permanently moved according to the movements of the drone as determined by the inertial unit thereof, and in the opposite direction of the displacement detected. In other words, a "virtual camera" is defined by extracting a particular zone from the hemispherical image, this zone being dynamically displaced in the hemispherical image in the opposite direction of the movements of the drone in order to annihilate the oscillations that would otherwise be observed on the image. Thus, in the case illustrated in FIG. 2b, where the drone plunges downward from a pitch angle δ (arrow 38) with respect to the vertical V, the capture window will be moved upwards (arrow 44) by an angle of the same value, thus reducing the central axis of the sector 36 of the "virtual camera" in the direction of the horizon HZ. As illustrated in the figures, as the movements of the drone forward are more frequent than those to the rear and that, on the other hand, the areas of interest (terrain overflown) are rather below the level of the drone above it, it may be advantageous to tilt down the main axis A of the fisheye lens (eg -20 °), so as to cover a greater number of configurations of evolution of the drone and to ensure that the sector 36 corresponding to the capture zone of the "virtual camera" always remains in the field 42 of the fisheye lens. FIG. 3 is a block diagram of the various servo control and steering control elements of the drone, as well as correction of the displacements of the image according to the technique of the invention. It will be noted that, although these diagrams are presented in the form of interconnected circuits, the implementation of the various functions is essentially software, this representation having only an illustrative character. In general, as shown in Figure 3, the control system involves several nested loops for the control of the horizontal speed, the angular speed of the attitude of the drone and the variations of altitude, automatically or under control. of the user. The most central loop is the angular velocity control loop 100, which uses, on the one hand, the signals supplied by gyrometers 102 and, on the other hand, a reference constituted by angular velocity instructions 104. applied at the input of an angular speed correction stage 106, which itself controls a motor control stage 110 in order to control separately the speed of the various motors to correct the angular velocity of the drone by the combined action of the rotors driven by these engines.

The angular velocity control loop 100 is embedded in an attitude control loop 112, which operates from the indications provided by the gyros 102 and by accelerometers 114. The data from these sensors are applied to a stage 118. which produces an estimate of the real attitude of the drone, applied to an attitude correction stage 120. This stage 120 compares the actual attitude of the drone with angle commands generated by a circuit 122 from commands directly applied by the user 124 and / or from data generated internally by the autopilot of the drone via the circuit 126 for horizontal speed correction. The possibly corrected instructions applied to the circuit 120 and compared to the actual attitude of the drone are transmitted by the circuit 120 to the circuit 104 to control the motors appropriately. Finally, a horizontal speed control loop 130 includes a vertical video camera 132 and a telemetric sensor 134 acting as an altimeter. A circuit 136 processes the images produced by the vertical camera 132, in combination with the signals of the accelerometer 114 and the attitude estimation circuit 118, to produce data for obtaining an estimate of the horizontal speeds according to the two axes of pitch and roll of the drone, by means of a circuit 138.

The estimated horizontal speeds are corrected by the vertical speed estimate given by a circuit 140 and by an estimate of the altitude value, given by the circuit 142 from the information of the telemetric sensor 134. For the control of vertical displacements of the drone, the user 124 applies commands to an attitude setting calculating circuit 144, which instructions are applied to a calculation circuit of ascending speed instructions V, 146 via the altitude correction circuit 148 receiving the estimated attitude value given by the circuit 142. The calculated climbing speed V is applied to a circuit 150 which compares this set speed with the corresponding speed estimated by the circuit 140 and modifies the control data accordingly. motors (circuit 108) by increasing or decreasing the speed of rotation simultaneously on all the motors so as to minimize the difference between speed asc ensionnelle setpoint and measured rate of climb.

As regards more specifically the implementation of the invention, the front video camera 14 delivers raw video data (pixel data) applied to a windowing circuit 152 ensuring the selection of the useful pixels in a capture area, of which the position depends on the attitude of the drone at a given moment, as determined by the inertial unit 154 (including the gyrometers 102, the accelerometers 114 and the attitude estimation circuit 118). The video data extracted from the capture zone are delivered to a circuit 156 for correcting the geometric distortions introduced by the fisheye lens, so as to produce rectified video data, themselves delivered to a transmitting circuit 158 ensuring the transmission of the video image remote control device held by the user, in particular for display on the screen of this remote control device and eventual recording of the video sequence. Figure 4 shows an example of a scene, as captured by the fisheye lens and detected on the camera sensor. As can be seen, this image I has very strong geometric distortions, inherent to the hemispherical or quasi-hemispherical coverage of the fisheye lens, which is rectified on the flat surface of the sensor.

Only part of the image I produced by the fisheye lens is used. This part is determined according to i) the direction in which the "virtual camera" is pointed, ii) its field of view (shown at 36 in Figures 2a and 2b) and iii) its ratio width / height. A "capture zone" ZC is thus defined containing raw pixel data including the "useful zone" ZU corresponding to the field of the virtual camera after compensation for geometric distortions introduced by the fisheye lens. It should be noted that it is not useful to capture all the pixels of the image I formed on the camera sensor, but only a fraction of them (for the capture zone ZC). For example, if you want to obtain a final image in HD quality (1920 x 1080 pixels, or 2 Mpixel for the useful area ZU), it is necessary to have a very high resolution fisheye image at the start. to be able to extract a good quality HD view regardless of the direction in which the virtual camera points, for example a 14 Mpixel resolution sensor (4608 x 3288 pixels). Under these conditions, if we transfer the entire image I for processing, this would correspond to a 14 Mpixel pixel data stream for each image, leading to a frame rate (framerate) of the order of 6 frames per second (fps) at this resolution, which would be insufficient for a fluid video sequence (imposing a framerate close to 30 fps). Therefore, only the pixel data of the capture zone ZC really needed is transferred, for example a capture window ZC of about 2 Mpixel, which can be retrieved at a rate of 30 fps without any particular difficulty. It is thus possible to choose a high resolution sensor while maintaining a high image rate. In FIG. 5, the various treatments performed on the pixel data of the capture zone ZC are illustrated to arrive at the final image, compensated for geometric distortions. From the pixel data transferred from the capture zone ZC (FIG. 5a), the process extracts the pixel data from the gross useful area ZUB (FIG. 5b) and applies to them a mesh of triangles (technique that is itself known), which will straighten the image by stretching each triangle to give a ZUR rectified useful image (Figure 5c) with rectified pixel data. In particular, the strongly curved horizontal lines of the fisheye image will be corrected to make them rectilinear and produce an image corresponding to a natural vision, devoid of geometric distortions.

With reference to FIGS. 6 to 10, the manner in which the capture window ZC is modified and moved according to the orientation of the virtual camera will now be described. The windowing operation in fact involves moving the capture area (pixel data acquisition window, transferred from the sensor to the processing circuits) during the transfer of the video stream, while maintaining a high image rate. FIG. 6 gives an example of an image I delivered by the fisheye lens of the camera, which, as illustrated in FIG. 7, can produce various images corrected at the ZUR center (c) after extraction of the capture zone and correction of the geometrical distortions, at the top ZUR (h), at the bottom ZUR (b), at the left ZUR (g) or at the right ZUR (d), all from the same original image L In the case of roll movements on the left or right , the image is rotated as shown in a) and b) in Figure 8. The image correction is not particularly difficult, as it is sufficient to provide a light capture area. expanded and applying, after transfer of the pixel data, an image correction by rotation in one direction or the other, this correction being without significant impact on the flow of pixel data transferred from the sensor to the processing circuit. On the other hand, the rotations of the drone around the pitch axis 22 (when the drone plunges forward or, on the contrary, rears) introduce relatively large displacements of the catch zone ZC, upwards or downwards. around a central position.

With a conventional configuration where the sensor is oriented in a "landscape" format, these rotations cause displacements of the capture zone parallel to the direction of the frame of the sensor, which has the consequence of introducing significant drops in the flow rate. transfer of the pixel data from the sensor to the processing circuits, with a significant risk of falling the framerate: the change of the scanning sequence of the sensor to extract the capture zone ZC can indeed lead, because of the slowdown the pixel data rate, at a loss of some images of the sequence, with a correlative decrease of framerate up to 50%.

However, the oscillations around the pitch axis are the most frequent (drone advance / retreat, acceleration / deceleration phases ...). Also, as illustrated in FIG. 9, to compensate for these displacements, it is decided to turn the sensor by 90 ° with respect to a conventional orientation, that is to say to place it in the "portrait" configuration, so as to favor displacements in a direction perpendicular to the direction of frame scan DB (the position of the scan lines has been schematized in e ,, e2, e3, e4 ...): therefore, the movements of the capture areas related to pitching movements (ZCh and ZCh around ZC0) will have no impact on the bit rate of the pixel data delivered by the sensor. In other words, in this configuration the sensor is oriented so that its frame scanning direction DB is parallel to the pitch axis 22 of the drone. FIG. 10 is a counterpart of FIG. 9 for a movement of the capture zone ZC consecutive to a movement of the drone around its yaw axis 26. If, as in the case of FIG. - Mera perpendicular to its "natural" positioning, these movements of the capture zone ZC will be oriented in the same direction as the scanning direction DB of the sensor. But since the amplitudes of these variations are much smaller than those corresponding to pitching motions, the impact on the bit rate of pixel data delivered by the sensor will be minimal, and the risk of losing images will be small. Figures 11, 12 and 13 illustrate the deformations of wobble and jelly type observable on the image of a checkerboard, and that should be compensated in addition to geometric distortions of the fisheye lens. The "jelly" effect, illustrated in Figure 1, appears in the presence of rotations of the drone of high amplitude but relatively low frequency. For example, in the case of a rotation of the drone of 100 ° / s and a video capture at 30 fps, between the top and the bottom of the image the drone will have turned of 100 x 1/30 = 3, 33 °, which corresponds to a displacement of several pixels on the image (about one tile of the checkerboard between the top and the bottom of the image in the example of Figure 12). These artifacts are particularly troublesome on moving images, with permanent and variable distortion of straight lines.

To overcome this phenomenon, it is possible to adapt to each line L of the image the processing for obtaining the useful area ZU during the step of reprojection and recovery of the capture zone ZC, this correction line by line to cancel the artifact introduced by the rapid rotation of the drone.

Figure 13 illustrates another type of "wobble effect" artefact, mainly caused by motor vibration: the wobble effect is due to high frequency and low amplitude oscillations caused by motor vibration. unlike the jelly effect, which is a low-frequency, high-amplitude effect caused by the drone's rotations to move. The wobble effect is mainly corrected by an appropriate mechanical damping of the camera support, allowing filtering of motor vibrations. The residues of this mechanical filtering can be eliminated in the same way as for the jelly effect, by using the measurements of the gyrometers and by applying the corrections line by line.

Claims (6)

REVENDICATIONS1. A rotary wing drone (10), comprising: - a camera (14), comprising a lens pointing in a fixed direction (A) with respect to the body of the drone as well as a digital sensor collecting the image (I) formed by the objective; and an inertial unit (154) capable of measuring the Euler angles (9, 0, tp) characterizing the attitude of the drone with respect to an absolute terrestrial reference, characterized in that the objective is a field objective hemispherical fisheye type, and in that the drone further comprises: - extractor means (152), receiving as input a selection signal defining the position of a capture zone (ZC) of reduced size over the extent of sensor, and outputting raw pixel data corresponding to the capture area; servo-control means (152), receiving as input at least one Euler angle delivered by the inertial unit (154), and capable of modifying the selection signal dynamically in a direction opposite to that of a change the attitude of the drone detected by the inertial unit and characterized by a corresponding variation of said at least one Euler angle; image processing means (156), receiving as input the raw pixel data delivered by the extractor means and outputting corrected pixel data, compensated for the geometric distortions introduced by the fisheye lens onto the image collected by the sensor in the region of the capture zone selected by the extracting means; and output means (158) for outputting the corrected pixel data for transmission to a video display or recording device. 2. The drone of claim 1, wherein the servo means are adapted to modify the selection signal so that the pixel data delivered by the extractor means correspond to an image centered on the horizon, or centered on an orientation fixed in relation to the horizon. 3. The drone of claim 1, wherein the camera is mounted in the drone so that the raster scanning direction (Dg) of the digital sensor is oriented parallel to the pitch axis (22) of the drone. 4. The drone of claim 1, wherein said at least one Euler angle is a pitch angle (q) of the drone, and the servo means are adapted to modify the selection signal so as to translate the capture zone in a first direction, parallel to a main axis of the sensor. 5. The drone of claim 4, wherein said at least one Euler angle is a yaw angle (tif) of the drone, and the servo means are adapted to modify the selection signal so as to translate the zone. capture in a second direction, perpendicular to said first direction. 6. The drone of claim 1, wherein the digital sensor is a scanning sensor delivering the line-by-line pixel data (,), and the image processing means (156) is able to apply to each line. (e,) of the sensor an additional correction to compensate for relative displacements of pixels from one line to the next induced by rotations of the drone around a yaw, pitch and / or roll axis.