Point cloud target tracking method, electronic device, medium and vehicle

Abstract

The present application relates to the technical field of automatic driving, and particularly provides a point cloud target tracking method, an electronic device, a medium and a vehicle, aiming to solve the technical problem that the existing target tracking method is difficult to simultaneously process two kinds of target inputs containing detection targets and segmentation targets. To this end, the point cloud target tracking method of the present application comprises: obtaining detection targets and segmentation targets respectively; matching the detection targets with first tracking trajectories to obtain first matching results; matching the segmentation targets with the first tracking trajectories that are not matched to obtain second matching results; matching the detection targets that are not matched and the segmentation targets that are not matched with second tracking trajectories respectively to obtain third matching results; and determining optimal state estimations of the detection targets and the segmentation targets based on the first matching results, the second matching results and the third matching results respectively. In this way, the matching accuracy of the detection targets and the segmentation targets is improved.

Description

Technical Field

[0001] This invention relates to the field of autonomous driving technology, specifically providing a point cloud target tracking method, electronic device, medium, and vehicle. Background Technology

[0002] LiDAR point cloud target tracking is one of the key technologies of autonomous driving perception module. By performing data association and state filtering on the target detection results of continuous frames, it can realize environmental perception in complex dynamic scenes and provide data support for subsequent decision planning modules.

[0003] Existing point cloud target tracking methods typically assume that the input target comes from a single detection algorithm. For example, the target detection module uses a deep learning model to output the target's 3D box information, including data such as target category, center point, orientation angle, and 3D dimensions. The tracking module first constructs a similarity matrix based on the distance between the detected target and the track center point, and uses the Hungarian algorithm to solve for the matching results. For a matched target, its observed state X = [x,y,z,l,w,h,r] (where x,y,z represent the target center point position, l,w,h represent the target's length, width, and height dimensions, and r represents the target's orientation angle) is extracted. Kalman filtering is then used to estimate the optimal state to obtain the smoothed target motion state.

[0004] Deep learning algorithms suffer from poor interpretability and difficulty in achieving 100% recall. Therefore, in practical engineering, to ensure driving safety, a traditional point cloud segmentation module is usually chained after the deep learning model, meaning that object detection employs two different methods. Traditional segmentation algorithms use point cloud geometric information for clustering, but cannot accurately estimate the target's center point, 3D dimensions, orientation angle, etc., and can only represent the target as a polygon enclosing the point cloud. Because the outputs of these two algorithms differ in shape, existing target tracking methods struggle to handle both types of target inputs simultaneously.

[0005] Accordingly, there is a need in this field for a new point cloud target tracking scheme to solve the above problems. Summary of the Invention

[0006] To overcome the aforementioned shortcomings, this invention is proposed to provide a solution, or at least a partial solution, to the technical problem that existing target tracking methods struggle to simultaneously handle both target inputs involving target detection and target segmentation. This invention provides a point cloud target tracking method, an electronic device, a medium, and a vehicle.

[0007] In a first aspect, the present invention provides a point cloud target tracking method, the method comprising: acquiring a detected target and a segmented target respectively, wherein the detected target is obtained based on a deep learning algorithm and the segmented target is obtained based on a point cloud segmentation algorithm; matching the detected target with a first tracking trajectory to obtain a first matching result; matching the segmented target with an unmatched first tracking trajectory to obtain a second matching result; matching the unmatched detected target and the unmatched segmented target with the second tracking trajectory respectively to obtain a third matching result; and determining optimal state estimates of the detected target and the segmented target based on the first matching result, the second matching result, and the third matching result respectively.

[0008] In one specific implementation, matching the detected target with a first tracking trajectory to obtain a first matching result includes: determining the Mahalanobis distance between the detected target and the first tracking trajectory; obtaining the first matching result based on the Mahalanobis distance; and / or

[0009] Matching the segmented target with the unmatched first tracking trajectory to obtain a second matching result includes: determining the Euclidean distance between the centroid of the segmented target and the centroid of the unmatched first tracking trajectory; obtaining the second matching result based on the Euclidean distance; and / or

[0010] The unmatched detected targets and unmatched segmented targets are matched with the second tracking trajectory to obtain a third matching result, including: determining the Euclidean distance between the centroid of the unmatched detected target and the second tracking trajectory and the Euclidean distance between the centroid of the unmatched segmented target and the centroid of the second tracking trajectory; and determining the third matching result based on the Euclidean distance.

[0011] In one specific implementation, determining the optimal state estimates of the detected target and the segmented target based on the first matching result, the second matching result, and the third matching result respectively includes:

[0012] Based on the first matching result, the motion state of the detected target is filtered to obtain the optimal state estimate of the detected target; and / or

[0013] Based on the second matching result, a 3D bounding box is fitted to the segmented target to obtain the corresponding 3D bounding box. The motion state of the segmented target is then filtered based on the 3D bounding box to obtain the optimal state estimate of the segmented target; and / or

[0014] Determine whether the target corresponding to the third matching result is a segmentation target; if so, perform 3D bounding box fitting on the segmentation target based on the third matching result to obtain the 3D bounding box corresponding to the segmentation target, and filter the motion state of the segmentation target based on the 3D bounding box to obtain the optimal state estimate of the segmentation target; if not, filter the motion state of the detection target based on the third matching result to obtain the optimal state estimate of the detection target.

[0015] In one specific implementation, fitting a 3D bounding box to the segmented target based on the second matching result to obtain a 3D bounding box corresponding to the segmented target includes: constructing a minimum bounding rectangle based on the segmented target and the second matching result; obtaining the vertex closest to the vehicle from the four vertices of the minimum bounding rectangle and using the closest vertex as a vertex of the 3D bounding box; and using the size predicted by the first tracking trajectory corresponding to the segmented target as the three-dimensional size of the 3D bounding box to obtain the 3D bounding box corresponding to the segmented target.

[0016] In one specific implementation, the method further includes: determining whether the disappearance time of the unmatched first tracking trajectory is greater than a preset time; if so, deleting the unmatched first tracking trajectory from the first tracking trajectory list.

[0017] Determine whether the disappearance time of the unmatched second tracking trajectory is greater than a preset time. If yes, delete the unmatched second tracking trajectory from the second tracking trajectory list. If no, determine whether the survival time of the unmatched second tracking trajectory is greater than a preset time. If the survival time of the unmatched second tracking trajectory is greater than the preset time, convert the unmatched second tracking trajectory into a first tracking trajectory.

[0018] In one specific implementation, the method further includes: for unmatched targets, generating a second tracking trajectory based on the observations of the unmatched targets.

[0019] In one specific implementation, before matching the detected target with the first tracking trajectory to obtain the first matching result, the method further includes: transforming the detected target and the segmented target from the vehicle coordinate system to the world coordinate system.

[0020] In a second aspect, an electronic device is provided, comprising a processor and a storage device adapted to store a plurality of program codes adapted to be loaded and executed by the processor to perform the point cloud target tracking method described in any of the preceding claims.

[0021] In a third aspect, a computer-readable storage medium is provided, wherein a plurality of program codes are stored therein, the program codes being adapted to be loaded and run by a processor to perform the point cloud target tracking method described in any of the preceding claims.

[0022] In a fourth aspect, a vehicle is provided, the vehicle including the aforementioned electronic equipment.

[0023] The above-described technical solutions of the present invention have at least one or more of the following beneficial effects:

[0024] This invention provides a point cloud target tracking method. For simultaneously received detected and segmented targets, a multi-stage matching approach is used to perform trajectory matching on both targets separately, determining the optimal state estimates for each target based on the matching results. This solves the problem of mismatching, improves target matching accuracy, enhances the stability and robustness of target motion state tracking, and ensures the driving safety of autonomous vehicles. Attached Figure Description

[0025] The disclosure of this invention will become more readily understood with reference to the accompanying drawings. It will be readily understood by those skilled in the art that these drawings are for illustrative purposes only and are not intended to limit the scope of protection of this invention. Furthermore, similar numbers in the drawings are used to denote similar components, wherein:

[0026] Figure 1 This is a schematic diagram of the main steps of a point cloud target tracking method according to an embodiment of the present invention;

[0027] Figure 2 This is a schematic diagram of a segmentation target 3D box fitting method according to an embodiment of the present invention;

[0028] Figure 3 This is a schematic diagram of a multi-stage matching method according to an embodiment of the present invention;

[0029] Figure 4 This is a schematic diagram of a track management process according to an embodiment of the present invention;

[0030] Figure 5 This is a schematic diagram of the overall process of a point cloud target tracking method according to an embodiment of the present invention:

[0031] Figure 6 This is a schematic diagram of the structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0032] Some embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0033] In the description of this invention, "module" and "processor" can include hardware, software, or a combination of both. A module can include hardware circuitry, various suitable sensors, communication ports, memory, and may also include software components, such as program code, or a combination of software and hardware. A processor can be a central processing unit, microprocessor, image processor, digital signal processor, or any other suitable processor. The processor has data and / or signal processing capabilities. The processor can be implemented in software, in hardware, or a combination of both. Non-transitory computer-readable storage media includes any suitable medium capable of storing program code, such as magnetic disks, hard disks, optical disks, flash memory, read-only memory, random access memory, etc. The term "A and / or B" means all possible combinations of A and B, such as only A, only B, or A and B. The terms "at least one A or B" or "at least one of A and B" have a similar meaning to "A and / or B" and can include only A, only B, or A and B. The singular terms "a" or "this" can also include plural forms.

[0034] To ensure driving safety, a traditional point cloud segmentation module is typically chained after a deep learning model, meaning that object detection employs two different methods. Traditional segmentation algorithms use point cloud geometry for clustering, but cannot accurately estimate the target's center point, 3D dimensions, orientation angle, etc., and can only represent the target as a polygon enclosing the point cloud. Because the outputs of these two algorithms differ in shape, existing target tracking methods struggle to simultaneously handle both detected and segmented target inputs. Furthermore, traditional segmentation algorithms often exhibit oversegmentation and undersegmentation; directly applying existing single-stage matching algorithms will result in numerous false matches, hindering subsequent state filtering.

[0035] To address this, this invention proposes a point cloud target tracking method that can simultaneously process detected and segmented targets. It employs a multi-stage matching approach, sequentially matching the detected and segmented targets with the track. Different similarity distances are used for matching targets with different shapes, effectively resolving issues such as inconsistent target shapes and mismatches. Before applying state filtering to the matched segmented target, historical information is used to fit a 3D box, estimating its center point, 3D dimensions, and orientation angle. Then, Kalman filtering is used to obtain the optimal state estimate. This solves the mismatch problem, improves target matching accuracy, enhances the stability and robustness of target motion state tracking, and ensures the driving safety of autonomous vehicles.

[0036] See appendix Figure 1 , Figure 1 This is a schematic diagram of the main steps of a point cloud target tracking method according to an embodiment of the present invention.

[0037] like Figure 1 As shown, the point cloud target tracking method in this embodiment of the invention mainly includes the following steps S101-S105.

[0038] Step S101: Obtain the detection target and the segmentation target respectively. The detection target is obtained based on a deep learning algorithm, and the segmentation target is obtained based on a point cloud segmentation algorithm.

[0039] The detected target refers to the target detected by deep learning algorithms. It is usually represented by a 3D bounding box, which includes information such as target category, center point, centroid, three-dimensional size, and orientation angle.

[0040] The segmentation target refers to the target obtained using traditional point cloud segmentation algorithms. It is usually represented by a convex polygon and includes information such as the list of convex polygon vertices and centroid.

[0041] The centroid mentioned above refers to the centroid of the laser point cloud surrounded by the target 3D box or convex polygon.

[0042] In one specific implementation, before matching the detected target with the first tracking trajectory to obtain the first matching result, the method further includes: transforming the detected target and the segmented target from the vehicle coordinate system to the world coordinate system.

[0043] Target detection and segmentation are performed in the vehicle coordinate system, while tracking is performed in the world coordinate system. Therefore, before multi-stage matching, the detected and segmented targets can be transformed from the vehicle coordinate system to the world coordinate system based on the vehicle's pose information. Pose information includes position and attitude, namely the vehicle's x-coordinate, y-coordinate, z-coordinate, pitch angle, roll angle, and yaw angle in the world coordinate system.

[0044] Step S102: Match the detected target with the first tracking trajectory to obtain the first matching result. The first tracking trajectory refers to several historical target trajectories maintained in the tracking algorithm.

[0045] Step S103: Match the segmented target with the unmatched first tracking trajectory to obtain the second matching result.

[0046] Step S104: Match the unmatched detected targets and unmatched segmented targets with the second tracking trajectory respectively to obtain the third matching result.

[0047] In one specific implementation, the method further includes: for unmatched targets, generating a second tracking trajectory based on the observations of the unmatched targets.

[0048] Unmatched targets refer to both unmatched detected targets and unmatched tracked targets. For unmatched targets, a new trajectory can be created based on the target's observed values ​​[x,y,z,l,w,h,r]. Because this trajectory appears only briefly and may be a false detection, it is called a candidate trajectory, or the second tracking trajectory.

[0049] Step S105: Determine the optimal state estimates of the detected target and the segmented target based on the first matching result, the second matching result and the third matching result, respectively.

[0050] Based on steps S101-S105 above, for simultaneously received detected and segmented targets, a multi-stage matching method is used to perform trajectory matching on the detected and segmented targets separately, so as to determine the optimal state estimates of the detected and segmented targets according to the matching results. This solves the problem of mismatching, improves the target matching accuracy, enhances the stability and robustness of target motion state tracking, and ensures the driving safety of autonomous vehicles.

[0051] In one specific implementation, matching the detected target with a first tracking trajectory to obtain a first matching result includes: determining the Mahalanobis distance between the detected target and the first tracking trajectory; obtaining the first matching result based on the Mahalanobis distance; and / or

[0052] The segmented target is matched with the unmatched first tracking trajectory to obtain a second matching result, including: determining the Euclidean distance between the centroid of the segmented target and the centroid of the unmatched first tracking trajectory; obtaining the second matching result based on the Euclidean distance; and / or

[0053] The unmatched detected targets and unmatched segmented targets are matched with the second tracking trajectory respectively to obtain a third matching result, including: determining the Euclidean distance between the centroid of the unmatched detected target and the second tracking trajectory and the Euclidean distance between the centroid of the unmatched segmented target and the centroid of the second tracking trajectory respectively; and determining the third matching result based on the Euclidean distance.

[0054] This application involves matching the detected target, segmented target, and track or candidate track in three stages.

[0055] The matching between the detected target and the first tracking track is a one-stage matching process. First, the Mahalanobis distance between the detected target and the track is calculated to construct a similarity matrix. Then, the Hungarian algorithm is used to solve for the first matching result.

[0056] The definition of Mahalanobis distance is: Where X t This represents the observed state of the target at time t, i.e., X. t =(x D ,y D ,z D ,l D ,w D ,h D ,r D ) T Y t Y represents the predicted state of the track at time t. t =(x T ,y T ,z T ,l T ,w T ,h T ,r T ) T Σ represents the covariance matrix of the target state filter.

[0057] In the two-stage matching process, the unmatched tracks from the first-stage matching are matched with the segmentation target. A similarity matrix is ​​constructed based on the Euclidean distance between the centroids of the two tracks, and the second matching result is obtained by solving the Hungarian algorithm.

[0058] The Euclidean distance is defined as: Where C S To divide the centroid of the target, C T Let be the centroid of the track.

[0059] In the three-stage matching process, the unmatched detected targets from the first stage and the unmatched segmented targets from the second stage are first merged, and then matched with the candidate track. Specifically, the Euclidean distance between the centroid of the unmatched detected target and the candidate track, and the Euclidean distance between the centroid of the unmatched segmented target and the centroid of the candidate track are determined. Then, the Hungarian algorithm is used to solve for these two Euclidean distances to obtain the third matching result.

[0060] In one specific implementation, determining the optimal state estimates of the detected target and the segmented target based on the first matching result, the second matching result, and the third matching result respectively includes: filtering the motion state of the detected target based on the first matching result to obtain the optimal state estimate of the detected target; and / or

[0061] Based on the second matching result, a 3D bounding box is fitted to the segmented target to obtain the corresponding 3D bounding box. The motion state of the segmented target is then filtered based on the 3D bounding box to obtain the optimal state estimate of the segmented target; and / or

[0062] Determine whether the target corresponding to the third matching result is the segmentation target; if so, perform 3D bounding box fitting on the segmentation target based on the third matching result to obtain the 3D bounding box corresponding to the segmentation target, and filter the motion state of the segmentation target based on the 3D bounding box to obtain the optimal state estimate of the segmentation target; if not, filter the motion state of the detected target based on the third matching result to obtain the optimal state estimate of the detected target.

[0063] 3D bounding box fitting is used to estimate the center point, 3D dimensions, and orientation angle information of segmented targets.

[0064] For a target that has been successfully matched in the first stage, since it already contains information on the center point, three-dimensional dimensions, and orientation angle, there is no need to perform 3D box fitting. The motion state of the target can be filtered directly based on the first matching result to obtain the optimal state estimate of the target.

[0065] For the segmented targets that are successfully matched in the second stage, 3D box fitting is required to obtain the 3D bounding box corresponding to the segmented target. The motion state of the segmented target is then filtered based on the 3D bounding box to obtain the optimal state estimate of the segmented target.

[0066] In addition, for targets that are successfully matched in the three stages, it is first determined whether the target is a segmentation target. If it is, 3D box fitting is performed to obtain the 3D bounding box corresponding to the segmentation target. The motion state of the segmentation target is then filtered based on the 3D bounding box to obtain the optimal state estimate of the segmentation target.

[0067] In one specific implementation, the 3D bounding box fitting of the segmented target based on the second matching result is used to obtain the 3D bounding box corresponding to the segmented target, including: constructing a minimum bounding rectangle based on the segmented target and the second matching result; obtaining the vertex closest to the vehicle from the four vertices of the minimum bounding rectangle and using the closest vertex as a vertex of the 3D bounding box; using the size predicted by the first tracking trajectory corresponding to the segmented target as the three-dimensional size of the 3D bounding box to obtain the 3D bounding box corresponding to the segmented target.

[0068] For example, such as Figure 2 As shown, a convex polygon represents a segmentation target, and the oblique arrow indicates the target orientation predicted by the track matching the segmentation target at the current moment. Based on the convex polygon and the target orientation, a minimum bounding rectangle can be constructed, obtaining the four vertices A, B, C, and D of the rectangle. The vertex closest to the vehicle (i.e., vertex A) is selected as a vertex of the 3D box. The length, width, and height predicted by the track matching the segmentation target are taken as the three-dimensional dimensions of the 3D box, thus obtaining the final 3D bounding box, i.e., rectangle AB′C′D′.

[0069] The 3D box fitting method can perform 3D fitting on the segmented target to obtain the final 3D bounding box, thus realizing the estimation of both the segmented target and its optimal state.

[0070] The optimal state of the target can be estimated by using a linear Kalman filter algorithm to filter the target's motion state.

[0071] The state of a Kalman filter is defined as X = (x, y, z, l, w, h, r, v) x ,v y ,v z ,v r ) T The observations of the Kalman filter are O = (x, y, z, l, w, h, r). T Where x, y, z represent the center point of the target in the world coordinate system, l, w, h represent the three-dimensional dimensions of the target, r represents the orientation angle of the target in the world coordinate system, and v x ,v y ,v z ,v r Let x, y, z, and r represent the velocities, respectively.

[0072] like Figure 3 As shown, the results of multi-stage matching have four possibilities: successfully matched target, unmatched mixed target, unmatched track, and unmatched candidate track. The target and mixed target include both detected targets and segmented targets.

[0073] In one specific implementation, the method further includes: determining whether the disappearance time of the unmatched first tracking trajectory is greater than a preset time; if so, deleting the unmatched first tracking trajectory from the first tracking trajectory list.

[0074] Determine whether the disappearance time of the unmatched second tracking trajectory is greater than a preset time. If yes, delete the unmatched second tracking trajectory from the second tracking trajectory list. If no, determine whether the survival time of the unmatched second tracking trajectory is greater than a preset time. If the survival time of the unmatched second tracking trajectory is greater than the preset time, convert the unmatched second tracking trajectory into the first tracking trajectory.

[0075] The track management module is responsible for generating candidate tracks, eliminating tracks and candidate tracks, and converting between candidate tracks. For example... Figure 4 As shown, for the above unmatched track, if its disappearance time is greater than 2 frames, it is considered that the target has disappeared and the target is removed from the track list.

[0076] For the unmatched candidate tracks mentioned above, if their disappearance time is greater than 2 frames, the target is considered to have disappeared and is removed from the candidate track list. For the remaining candidate tracks, if their survival time is greater than 2 frames, they are converted into tracks and added to the track list for output.

[0077] For the aforementioned unmatched mixed targets, create new candidate tracks based on the target observations and add them to the candidate track list.

[0078] The track management module manages tracks and candidate tracks in real time, providing fundamental support for matching detected and segmented targets.

[0079] In one specific implementation, such as Figure 5 As shown, the acquired detection targets and segmented targets can be preprocessed first. Preprocessing refers to transforming the detection targets and segmented targets from the vehicle coordinate system to the world coordinate system.

[0080] Next, multi-stage matching is performed. For successfully matched detected targets, target state filtering is directly applied to obtain the target tracking result corresponding to the detected target. For successfully matched segmented targets, 3D box fitting is first performed to obtain 3D bounding boxes, and target state filtering is then applied based on these bounding boxes to obtain the target tracking result corresponding to the segmented target.

[0081] In addition, track management can be performed on unmatched tracks and unmatched candidate tracks.

[0082] Thus, compared with existing tracking methods, this invention employs multi-stage matching and 3D box fitting technology, which can simultaneously receive the target outputs of two different detection algorithms, effectively improving the overall recall rate of the laser sensing module and ensuring the driving safety of autonomous vehicles. In addition, the use of multi-stage matching effectively reduces mismatch phenomenon and improves the stability and robustness of target motion state tracking.

[0083] It should be noted that although the steps in the above embodiments are described in a specific order, those skilled in the art will understand that in order to achieve the effects of the present invention, different steps do not necessarily have to be executed in such an order. They can be executed simultaneously (in parallel) or in other orders, and these variations are all within the scope of protection of the present invention.

[0084] Those skilled in the art will understand that all or part of the processes in the method of the above embodiment of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable storage medium can include any entity or device capable of carrying the computer program code, a medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory, a random access memory, an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc. It should be noted that the content included in the computer-readable storage medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable storage medium does not include electrical carrier signals and telecommunication signals.

[0085] Furthermore, the present invention also provides an electronic device. In one embodiment of the electronic device according to the present invention, such as Figure 6 As shown, the electronic device includes a processor 601 and a storage device 602. The storage device can be configured to store a program for executing the point cloud target tracking method of the above-described method embodiments. The processor can be configured to execute the program in the storage device, which includes, but is not limited to, a program for executing the point cloud target tracking method of the above-described method embodiments. For ease of explanation, only the parts related to the embodiments of the present invention are shown. For specific technical details not disclosed, please refer to the method section of the embodiments of the present invention.

[0086] Furthermore, the present invention also provides a computer-readable storage medium. In one embodiment of the computer-readable storage medium according to the present invention, the computer-readable storage medium can be configured to store a program for performing the point cloud target tracking method of the above-described method embodiments. This program can be loaded and run by a processor to implement the above-described point cloud target tracking method. For ease of explanation, only the parts related to the embodiments of the present invention are shown; for specific technical details not disclosed, please refer to the method section of the embodiments of the present invention. The computer-readable storage medium can be a storage device device comprising various electronic devices. Optionally, in the embodiments of the present invention, the computer-readable storage medium is a non-transitory computer-readable storage medium.

[0087] Furthermore, the present invention also provides a vehicle that includes the aforementioned electronic equipment.

[0088] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. A point cloud target tracking method, characterized in that, The method includes: The detection target and the segmentation target are obtained separately. The detection target is obtained based on a deep learning algorithm, and the segmentation target is obtained based on a point cloud segmentation algorithm. The detected target is matched with the first tracking trajectory to obtain the first matching result. The first tracking trajectory refers to several historical target trajectories maintained in the tracking algorithm. The segmented target is matched with the unmatched first tracking trajectory to obtain a second matching result; The unmatched detected targets and unmatched segmented targets are matched with the second tracking trajectory to obtain the third matching result. The second tracking trajectory is generated based on the observations of the unmatched detected targets and unmatched tracked targets. The optimal state estimates of the detected target and the segmented target are determined based on the first matching result, the second matching result, and the third matching result, respectively.

2. The point cloud target tracking method according to claim 1, characterized in that, Matching the detected target with the first tracking trajectory to obtain a first matching result includes: determining the Mahalanobis distance between the detected target and the first tracking trajectory; obtaining the first matching result based on the Mahalanobis distance; and / or Matching the segmented target with the unmatched first tracking trajectory to obtain a second matching result includes: determining the Euclidean distance between the centroid of the segmented target and the centroid of the unmatched first tracking trajectory; obtaining the second matching result based on the Euclidean distance; and / or The unmatched detected targets and unmatched segmented targets are matched with the second tracking trajectory to obtain a third matching result, including: determining the Euclidean distance between the centroid of the unmatched detected target and the second tracking trajectory and the Euclidean distance between the centroid of the unmatched segmented target and the centroid of the second tracking trajectory; and determining the third matching result based on the Euclidean distance.

3. The point cloud target tracking method according to claim 1, characterized in that, Determining the optimal state estimates of the detected target and the segmented target based on the first matching result, the second matching result, and the third matching result respectively includes: Based on the first matching result, the motion state of the detected target is filtered to obtain the optimal state estimate of the detected target; and / or Based on the second matching result, a 3D bounding box is fitted to the segmented target to obtain the corresponding 3D bounding box. The motion state of the segmented target is then filtered based on the 3D bounding box to obtain the optimal state estimate of the segmented target; and / or Determine whether the target corresponding to the third matching result is a segmentation target; if so, perform 3D bounding box fitting on the segmentation target based on the third matching result to obtain the 3D bounding box corresponding to the segmentation target, and filter the motion state of the segmentation target based on the 3D bounding box to obtain the optimal state estimate of the segmentation target; if not, filter the motion state of the detected target based on the third matching result to obtain the optimal state estimate of the detected target.

4. The point cloud target tracking method according to claim 3, characterized in that, Based on the second matching result, a 3D bounding box is fitted to the segmented target to obtain the 3D bounding box corresponding to the segmented target, including: Construct a minimum bounding rectangle based on the segmentation target and the second matching result; From the four vertices of the minimum bounding rectangle, obtain the vertex closest to the vehicle, and use the closest vertex as a vertex of the 3D bounding box; The size predicted by the first tracking trajectory corresponding to the segmented target is used as the three-dimensional size of the 3D bounding box to obtain the 3D bounding box corresponding to the segmented target.

5. The point cloud target tracking method according to claim 1, characterized in that, Also includes: Determine whether the disappearance time of the unmatched first tracking trajectory is greater than a preset time. If so, delete the unmatched first tracking trajectory from the first tracking trajectory list. Determine whether the disappearance time of the unmatched second tracking trajectory is greater than a preset time. If yes, delete the unmatched second tracking trajectory from the second tracking trajectory list. If no, determine whether the survival time of the unmatched second tracking trajectory is greater than a preset time. If the survival time of the unmatched second tracking trajectory is greater than the preset time, convert the unmatched second tracking trajectory into a first tracking trajectory.

6. The point cloud target tracking method according to claim 1, characterized in that, Also includes: For unmatched targets, a second tracking trajectory is generated based on the observations of the unmatched targets.

7. The point cloud target tracking method according to claim 1, characterized in that, Before matching the detected target with the first tracking trajectory to obtain the first matching result, the method further includes: transforming the detected target and the segmented target from the vehicle coordinate system to the world coordinate system.

8. An electronic device comprising a processor and a storage device, said storage device being adapted to store a plurality of program codes, characterized in that, The program code is adapted to be loaded and run by the processor to perform the point cloud target tracking method according to any one of claims 1 to 7.

9. A computer-readable storage medium storing a plurality of program codes, characterized in that, The program code is adapted to be loaded and run by a processor to perform the point cloud target tracking method according to any one of claims 1 to 7.

10. A vehicle, characterized in that, The vehicle includes the electronic equipment as described in claim 8.

Images