JP6657934B2 - Object detection device - Google Patents

Description

  The present disclosure relates to an object detection device.

  Patent Literature 1 discloses an apparatus that improves the recognition accuracy of a tracking target by combining measurement points of reflection by an object in a plurality of frames when the number of measurement points of reflection by an object measured by a laser sensor is small. I have.

JP 2011-191227 A

  Incidentally, the measurement points measured by the laser sensor include, in addition to the measurement points of the reflection by the object, the measurement points of the reflection by the road surface. For example, when there is a road surface paint such as a lane boundary line, a measurement point of reflection by the road surface is acquired. In the device described in Patent Literature 1, when a synthesis target includes a measurement point on a road surface, the road surface may be erroneously detected as a tracking target.

  In the present technical field, it is desired to suppress erroneous detection of a road surface as a tracking target.

  An object detection device according to the present invention is an object detection device that outputs an object candidate that is a candidate for an object based on a plurality of measurement points measured by a laser sensor, and the position of the measurement point measured by the laser sensor Or, based on the reflection intensity, a measurement point, an object determination point that is a measurement point that measured an object on the road surface, or a point group classification unit that classifies the road surface determination point that is a measurement point that measured the road surface A first determination unit that clusters the plurality of object determination points into one or a plurality of first clusters based on the positions of the plurality of object determination points, and a road surface determination in which a distance between the positions of the plurality of object determination points is equal to or less than a predetermined value. Based on the position of the point and the position of the object determination point, a second determination for clustering a road surface determination point and an object determination point whose distance to the position of the object determination point is equal to or less than a predetermined value into one or a plurality of second clusters. Department and multiple A third determining unit that clusters the plurality of road surface determination points into one or a plurality of third clusters based on the position of the road surface determination point, and associates the first cluster with the object candidate output in the previous process. A correspondence unit that associates an object candidate that cannot be associated with the first cluster with the second cluster, and associates an object candidate that cannot be associated with the first cluster and the second cluster with the third cluster. An output unit that outputs the attached first cluster, second cluster, and third cluster as object candidates.

  Advantageous Effects of Invention According to the present invention, it is possible to suppress erroneous detection of a road surface as a tracking target while improving the ratio of correct detection of the tracking target.

1 is a block diagram illustrating a configuration of an object detection device according to an embodiment. It is a figure showing an example of an object judgment point and a road surface judgment point. FIG. 4 is a diagram illustrating an example of a cluster based on an object determination point. It is a figure showing an example of a cluster based on a road surface judgment point. FIG. 4 is an explanatory diagram of conditional clustering. It is explanatory drawing of the modification of conditional clustering. FIG. 4 is a diagram illustrating an example of a cluster by conditional clustering. It is a figure showing a cluster and a known object candidate. FIG. 4 is a diagram illustrating conventional clustering.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings. In the following description, the same or corresponding elements will be denoted by the same reference symbols, without redundant description.

  FIG. 1 is a block diagram illustrating a configuration of an object detection device 1 according to the embodiment. The object detection device 1 is mounted on a vehicle (a passenger car or the like) including the laser sensor 10. The object detection device 1 outputs an object candidate that is a candidate for an object based on a plurality of measurement points measured by the laser sensor 10.

  The laser sensor 10 is a detector that detects a surrounding environment of the vehicle. The surrounding environment includes a road surface and an object. The object is a tangible object, and includes a stationary object, a moving object, and the like. The laser sensor 10 includes an electromagnetic wave transmitting circuit, a receiving circuit, an arithmetic circuit for performing various arithmetic operations, and a storage unit. The electromagnetic wave is, for example, a laser beam or a millimeter wave. The laser sensor 10 emits an electromagnetic wave over a range of a set detection angle around the vehicle, and receives a reflected wave reflected by an object or a road surface. The laser sensor 10 detects a position of an object or a road surface based on a detection angle (reception angle) of a reflected wave and a detection distance obtained from a propagation time of an electromagnetic wave required from transmission to reception. The position is a three-dimensional position and includes a height. Note that the detection distance may be obtained from the intensity of the reflected wave. The laser sensor 10 acquires a plurality of positions at which the electromagnetic waves are reflected as measurement points for each transmission. That is, the measurement point is an observation point at a position where the electromagnetic wave is reflected.

  The object detection device 1 includes an ECU (Electronic Control Unit) 11. The laser sensor 10 and the ECU 11 are each connected to a network that communicates using a CAN (Controller Area Network) communication circuit, and can perform mutual communication.

  The ECU 11 is an electronic control unit having a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), a CAN communication circuit, and the like. The ECU 11 controls hardware based on a signal output from the CPU, and realizes functions of components described later. As an example of a more specific operation, the ECU 11 operates the CAN communication circuit to input and output data, stores the input data in the RAM, loads the program stored in the ROM into the RAM, and loads the program into the RAM. Executes the specified program.

  The ECU 11 includes a point cloud classification unit 20, a clustering unit 30, a correspondence unit 400, and a tracking candidate output unit (output unit) 50.

  The point cloud classification unit 20 measures the measurement point based on the position or the reflection intensity of the measurement point measured by the laser sensor 10, an object determination point that is a measurement point obtained by measuring an object on the road surface, or measures the road surface. It is classified into any of the road surface determination points that are measurement points. The object determination point is a point estimated as a measurement point due to reflection from the object. The road surface determination point is a point estimated as a measurement point due to reflection from the road surface. In these determinations, when the reflection intensity is unstable, for example, a measurement point due to reflection from the road surface may be estimated as an object determination point, or a measurement point due to reflection from an object may be estimated as a road surface determination point. .

  The point group classification unit 20 classifies the measurement points whose height information measured by the laser sensor 10 is equal to or more than the first threshold as object determination points. The first threshold is a threshold for classifying an object or a road surface based on height information. The point cloud classification unit 20 classifies the measurement point whose height information measured by the laser sensor 10 is not equal to or greater than the first threshold value as a road surface determination point. The point cloud classification unit 20 may classify the object into either an object determination point or a road surface determination point by combining the height information and the reflection intensity. For example, the point cloud classification unit 20 determines a measurement point where the height information measured by the laser sensor 10 is included in a predetermined range including the first threshold and the reflection intensity is equal to or less than the second threshold as a road surface determination point. It may be classified.

  FIG. 2 is a diagram illustrating an example of an object determination point and a road surface determination point. FIG. 2 shows a forward image G1 from a camera sensor (not shown) and a map image G2 subjected to coordinate transformation so that the surroundings of the vehicle are overlooked. FIG. 2 illustrates a case where one preceding vehicle and two following vehicles exist. As shown in FIG. 2, the map image G2 shows an object determination point 21 (white circle) and a road surface determination point 22 (black circle). On the map image G2, the object determination point 21 is concentrated at a position corresponding to the preceding vehicle H shown in the front image G1. On the map image G2, the road surface determination points 22 are concentrated at positions corresponding to the lane boundary lines L shown in the front image G1. The reflection from the road surface becomes strong or disturbed at the lane boundary and the road surface paint. For this reason, the road surface determination point 22 and the object determination point 21 are mixed at a position corresponding to the lane boundary line L shown in the front image G1. Further, an example is shown in which the measurement points of the two following vehicles are insufficient and cannot be accurately confirmed in the map image G2.

  The clustering unit 30 includes an object clustering unit (first determination unit) 301, a condition clustering unit (second determination unit) 302, and a road surface clustering unit (third determination unit) 303. The object clustering unit 301 clusters the plurality of object determination points 21 into one or a plurality of first clusters C1 based on the positions of the plurality of object determination points 21. A cluster is a collective term for a plurality of measurement points and is a group. Clustering refers to grouping a plurality of measurement points. The object clustering unit 301 extracts the adjacent object determination points 21 as the same first cluster. As a specific example, the object clustering unit 301 projects the object determination point 21 on a two-dimensional plane. Next, the object clustering unit 301 converts the projected point group into a cluster grid map. Next, the object clustering unit 301 searches each column of the cluster grid map and assigns the same cluster number to the connected grids. Thereby, the object clustering unit 301 clusters the object determination points 21 corresponding to the connected grids into the same first cluster.

  FIG. 3 is a diagram illustrating an example of the first cluster C1 based on the object determination point 21. FIG. 3 shows a result of clustering the object determination point 21 and the road surface determination point 22 shown in FIG. 2 into the first cluster C1. As shown in FIG. 3, the approaching object determination points 21 are clustered into one first cluster C1.

  The road surface clustering unit 303 clusters the plurality of road surface determination points 22 into one or a plurality of third clusters C3 based on the positions of the plurality of road surface determination points 22. The clustering method of the road surface clustering unit 303 is the same as the clustering method of the object clustering unit 301 except that the clustering target is the road surface determination point 22.

  FIG. 4 is a diagram illustrating an example of the third cluster C3 based on the road surface determination point 22. FIG. 4 shows the result of clustering the object determination point 21 and the road surface determination point 22 shown in FIG. 2 into the third cluster C3. As shown in FIG. 4, the adjacent road surface determination points 22 are clustered into one third cluster C3.

  The condition clustering unit 302 determines that the distance to the position of the object determination point 21 is predetermined based on the position of the road surface determination point 22 and the position of the object determination point 21 whose distance to the position of the object determination point 21 is equal to or less than a predetermined value. The road surface determination point 22 and the object determination point 21 that are equal to or less than the value are clustered into one or a plurality of second clusters C2. As shown in FIG. 4, when all the road surface determination points 22 are clustered, an actual road surface or a lane boundary is also a tracking candidate, so that the possibility of erroneous detection increases. As an example, the condition clustering unit 302 limits the road surface determination point 22 to be clustered to only the road surface determination point 22 near the object determination point 21. FIG. 5 is an explanatory diagram of conditional clustering. The distance L1 between the road surface determination point 22 (22A) and the object determination point 21 is equal to or less than a predetermined value. The distance L2 between the road surface determination point 22 (22B) and the object determination point 21 is larger than a predetermined value. As shown in FIG. 5, among the road surface determination points 22, only the road surface determination point 22 (22A) whose distance from the position of the object determination point 21 is equal to or less than a predetermined value is subjected to clustering, and the position of the object determination point 21 is determined. The road surface determination point 22 (22B) in which the distance from the road surface does not become a predetermined value or less is not a target of clustering. The condition clustering unit 302 clusters the target road surface determination point 22 and the object determination point 21 as a second cluster C2 (C21). The clustering method of the condition clustering unit 302 is the same as the clustering method of the object clustering unit 301 except that the target road surface determination point 22 and the object determination point 21 are clustered.

  FIG. 6 is an explanatory diagram of a modified example of the conditional clustering. The condition clustering unit 302 may further cluster clusters using a time-series tracking model. The time-series tracking model is a model that estimates the position of the object candidate in the current process using the size (length and width) of the position, speed, and direction of the object candidate output in the previous process. FIG. 6A shows the measurement result of one vehicle, in which the left and right ends are the object determination points 21, and the center part is the road surface determination point 22. As shown in FIG. 6A, the condition clustering unit 302 clusters the road determination point 22 near the object determination point 21 and the object determination point 21 into a second cluster C2 by the above-described method. . In this case, even if it is actually one vehicle, it is separated into two clusters. The condition clustering unit 302 estimates the object candidate B in the current process based on the position, speed, direction, and size (length and width) of the object candidate, and clusters the object candidate B in a range where the object candidate B exists in the current process. Is included, clustering is performed as one second cluster C2 (C22).

  FIG. 7 is a diagram illustrating an example of the second cluster C2 based on conditional clustering. FIG. 6 shows a result of clustering the object determination point 21 and the road surface determination point 22 shown in FIG. 2 into the second cluster C2. As shown in FIG. 6, the road surface determination point 22 and the object determination point 21 whose distance from the position of the object determination point 21 is equal to or less than a predetermined value are clustered into one or a plurality of second clusters C2 (C21). . Thus, as compared with FIG. 4, unnecessary clusters that cause erroneous detection are reduced, and clustering can be performed even when the object determination point 21 and the road surface determination point 22 coexist and cannot be clustered conventionally. Further, by applying a modified example of conditional clustering, it is possible to prevent one vehicle from being clustered into two (second cluster C2 (C22)). For this reason, it can be avoided that the correspondence unit 400 described later mistakenly associates.

  Further, the clustering unit 30 sets a certainty factor for each cluster for the processing of the corresponding unit 400 at the subsequent stage. The clustering unit 30 sets the certainty factor of the first cluster C1 higher than the certainty factors of the second cluster C2 and the third cluster C3 (high certainty factor). The clustering unit 30 sets the certainty factor of the second cluster C2 lower than the certainty factor of the first cluster C1 and higher than the certainty factor of the third cluster C3 (medium). That is, the clustering unit 30 sets the certainty factor of the third cluster C3 to be lower than the certainty factors of the first cluster C1 and the second cluster C2 (low certainty factor). As described above, the clustering unit 30 sets the certainty in the order of the first cluster C1, the second cluster C2, and the third cluster C3 so that the certainty decreases.

  The correspondence unit 400 associates the first cluster C1 with the object candidate output in the previous processing, associates an object candidate that cannot be associated with the first cluster C1 with the second cluster C2, and associates the first cluster C1 The object candidate that cannot be associated with the second cluster C2 is associated with the third cluster C3. The correspondence unit 400 performs the above-described processing based on the certainty factor set for each cluster.

  The associating unit 400 associates objects with candidate objects in descending order of certainty factor. The corresponding unit 400 includes a first corresponding unit 40, a second corresponding unit 41, and a third corresponding unit 42. The first correspondence unit 40 first associates the first cluster C1 (high confidence) with the object candidate output in the previous process. The association is performed by a known method using, for example, a Euclidean distance or a Mahalanobis distance. The first corresponding unit 40 outputs information on the associated tracking candidate and the first cluster C1 to the tracking candidate output unit 50. The first corresponding unit 40 outputs to the tracking candidate output unit 50 the first cluster C1 that cannot be associated with all the tracking candidates. When the association determination is completed for all the first clusters C <b> 1, the first corresponding unit 40 outputs the uncorrelated tracking candidates to the second corresponding unit 41.

  The second corresponding unit 41 associates the second cluster C2 (medium confidence level) with a tracking candidate that the first corresponding unit 40 did not associate. The second correspondence unit 41 outputs information on the associated tracking candidate and the second cluster C2 to the tracking candidate output unit 50. Further, the second correspondence unit 41 outputs the second cluster C2 that cannot be associated with all the tracking candidates to the tracking candidate output unit 50. When the association determination is completed for all the second clusters C2, the second corresponding unit 41 outputs the uncorrelated tracking candidates to the third corresponding unit 42.

  The third corresponding unit 42 associates the third cluster C3 (low confidence) with the tracking candidate that the first corresponding unit 40 and the second corresponding unit 41 did not associate. The third correspondence unit 42 outputs information on the associated tracking candidate and the third cluster C3 to the tracking candidate output unit 50. When the association determination has been completed for all the third clusters C3, the third corresponding unit 42 discards the unassociated clusters.

  The tracking candidate output unit 50 outputs the first cluster C1, the second cluster, and the third cluster associated by the association unit 400 as object candidates. In addition, the tracking candidate output unit 50 outputs the first cluster C1 and the second cluster that are not associated with each other as new object candidates.

  FIG. 8 is a diagram illustrating clusters and known object candidates. As shown in FIG. 8, at a certain time t, it is assumed that clusters C <b> 10 to C <b> 60 with certainty factors are observed around the own vehicle 2. In the previous process (time t-1), it is assumed that the tracking candidates T1 to T3 are vehicles.

  The correspondence between the tracking candidate T1 at time t-1 and the clusters C10 and C20 at time t will be described. Cluster C10 is the first cluster (high confidence). Cluster C20 is a second cluster (low confidence). The cluster C20 is closer in Euclidean distance. However, the tracking candidate T1 is associated with the cluster C10 in consideration of the certainty factor.

  The following describes how the tracking candidate T2 at time t-1 is associated with the cluster C30 and the cluster C40 at time t. Cluster C30 is the third cluster (medium confidence). Cluster C40 is the second cluster (low confidence). The cluster C40 is closer in Euclidean distance. However, the tracking candidate T2 is associated with the cluster C30 in consideration of the certainty factor.

  The following describes how the tracking candidate T3 at time t-1 is associated with the cluster C50 and the cluster C60 at time t. Cluster C50 and cluster C60 are the second cluster (low confidence). Since there is no cluster with high confidence in the surroundings, the tracking candidate T2 is associated with the cluster C50 having a short Euclidean distance, and the cluster C60 is discarded.

  As described above, according to the object detection device 1 according to the present embodiment, the first cluster C1, the second cluster C2, and the third cluster C3 correspond to the object candidates output in the previous process in the order of higher confidence. Attached. For this reason, the possibility that a cluster with low confidence is associated with a tracking candidate is reduced. Therefore, erroneous detection of a road surface as a tracking target can be suppressed. As a result, the correct detection ratio of the tracking target can be improved.

  Further, according to the object detection device 1 according to the present embodiment, it is possible to stably track an object that cannot be stably tracked in the past. First, conventional clustering will be described in order to compare with the effect of the object detection device 1 according to the present embodiment. FIG. 9 is a diagram for explaining conventional clustering. In FIG. 9, the measurement points of the rear vehicle H1 from time t to t3 are shown in plan view. In the conventional clustering, when it is determined as the road surface determination point 22, the clustering is not performed. Therefore, even when measuring the same target like the measurement points at times t, t + 1, and t + 3, only the minute clusters a to d are detected, or the measurement points at time t + 2 are not detected. In some cases, clusters were not measured at all. For this reason, there is a possibility that it is difficult to associate the target with the tracking target or the tracking target is lost.

  On the other hand, according to the object detection device 1 according to the present embodiment, the measurement points of the laser sensor 10 are clustered into the first cluster C1, the second cluster C2, and the third cluster C3 by the clustering unit 30. By employing the third cluster C3 obtained by clustering the object determination point 21 and the road surface determination point 22, the third cluster C3 can be generated from the minute clusters a and b at the time t and can be set as a tracking target. As a result, an object that could not be tracked stably in the past can be tracked stably.

  Furthermore, according to the object detection device 1 according to the present embodiment, compared to the conventional case of combining a plurality of frames, the processing can be accelerated by not performing the combining processing. For this reason, a temporal detection delay can be suppressed. When the sparse laser sensor 10 is mounted at a low mounting position of the vehicle, it may be difficult to classify the object determination point 21 and the road surface determination point 22. Even in such a case, the object detection device 1 according to the present embodiment can appropriately output the tracking target by employing the third cluster C3 in which the object determination points 21 and the road surface determination points 22 are clustered. Can be.

  The present invention can be implemented in various forms with various changes and improvements based on the knowledge of those skilled in the art based on the above-described embodiments.

  DESCRIPTION OF SYMBOLS 1 ... Object detection apparatus, 2 ... own vehicle, 10 ... Laser sensor, 20 ... Point group classification | category part, 21 ... Object determination point, 22 ... Road surface determination point, 30 ... Clustering part, 40 ... 1st corresponding part, 41 ... No. 2 corresponding part, 42 ... third corresponding part, 50 ... tracking candidate output part, 301 ... object clustering part (first determination part), 302 ... condition clustering part (second determination part), 303 ... road surface clustering part (third part) Judging part), 400 ... corresponding part.

Claims (1)

An object detection device that outputs an object candidate that is a candidate for an object based on a plurality of measurement points measured by a laser sensor,
Based on the position or reflection intensity of the measurement point measured by the laser sensor, the measurement point, the object determination point that is a measurement point that measured an object on the road surface, or a road surface that is a measurement point that measured the road surface A point group classifying unit that classifies any of the judgment points,
A first determination unit that clusters the plurality of object determination points into one or a plurality of first clusters based on positions of the plurality of object determination points;
The road surface where the distance to the position of the object determination point is equal to or less than a predetermined value, based on the position of the road surface determination point where the distance to the position of the object determination point is equal to or less than a predetermined value, and A second determination unit that clusters the determination point and the object determination point into one or more second clusters;
A third determination unit configured to cluster the plurality of road surface determination points into one or a plurality of third clusters based on positions of the plurality of road surface determination points;
It said first and associates the previous object candidate is a cluster and object candidate output in the last processing, and associates the previous object candidate that can not be associated with the first cluster and the second cluster, the first A corresponding unit that associates the previous object candidate that cannot be associated with one cluster and the second cluster with the third cluster;
The first cluster associated by the corresponding unit, and outputs the second cluster and the third cluster as the object candidate, the first cluster and the second cluster that is not associated with the corresponding portion An output unit that outputs as the new object candidate ,
An object detection device comprising:

Images