Multi-target tracking method and device, storage medium and electronic equipment

Abstract

The embodiment of the invention provides a multi-target tracking method and device, a storage medium and electronic equipment, and the method comprises the steps: carrying out the matching of each to-be-matched track and an observation target according to a three-dimensional matching parameter, and obtaining a three-dimensional matching probability corresponding to each to-be-matched track; matching each to-be-matched track with the observation target according to the two-dimensional matching parameters to obtain a two-dimensional matching probability corresponding to each to-be-matched track, the two-dimensional matching parameters are used for evaluating the matching degree between the two-dimensional projection corresponding to each track to be matched on a specified two-dimensional plane perpendicular to the horizontal plane and the two-dimensional projection of the observation target; fusing the three-dimensional matching probability corresponding to each track to be matched and the two-dimensional matching probability corresponding to each track to be matched into a fusion matching probability corresponding to each track to be matched; and when the observation target is determined to be matched with the target tracking trajectory based on the fusion matching probability corresponding to each to-be-matched trajectory, updating the target tracking trajectory by using the observation target.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of autonomous driving, and specifically, to a multi-target tracking method and device, a storage medium, and an electronic device. Background Art

[0002] In the field of autonomous driving, in order to ensure the safety and reliability of vehicle operation, multi-target tracking is required. In multi-target tracking, association matching generally uses the Hungarian algorithm, which comprehensively considers the weights (similarity) between multiple targets and determines the matching relationship between targets. However, if the distance between different sensor observations of the same target is far, they generally cannot be associated and matched together, so that the observations of the two sensors cannot be fused, thereby reducing the accuracy of multi-target tracking. It can be seen that the multi-target tracking method in the related art has the problem of poor target tracking accuracy. Summary of the invention

[0003] The embodiments of the present application provide a multi-target tracking method and device, a storage medium and an electronic device to at least solve the problem of poor target tracking accuracy in the multi-target tracking method in the related art.

[0004] According to one aspect of an embodiment of the present application, a multi-target tracking method is provided, including: determining a group of to-be-matched trajectories corresponding to observed targets, wherein each to-be-matched trajectory in the group of to-be-matched trajectories is a tracking trajectory of a tracking target of an autonomous driving vehicle, and the observed target is a target observed by the autonomous driving vehicle at a target time; matching each to-be-matched trajectory with the observed target according to three-dimensional matching parameters to obtain a three-dimensional matching probability corresponding to each to-be-matched trajectory, wherein the three-dimensional matching parameters are used to evaluate the degree of matching between each to-be-matched trajectory and the observed target in three-dimensional space; matching each to-be-matched trajectory with the observed target according to two-dimensional matching parameters. The method comprises the steps of: matching the observed target with the target tracking track in the set of trajectories to be matched, wherein the two-dimensional matching parameters are used to evaluate the degree of matching between the two-dimensional projection corresponding to each trajectory to be matched and the two-dimensional projection of the observed target on a specified two-dimensional plane; fusing the three-dimensional matching probability corresponding to each trajectory to be matched and the two-dimensional matching probability corresponding to each trajectory to be matched to obtain a fused matching probability corresponding to each trajectory to be matched; and updating the target tracking trajectory using the observed target when it is determined that the observed target matches the target tracking trajectory in the set of trajectories to be matched based on the fused matching probability corresponding to each trajectory to be matched.

[0005] According to another aspect of an embodiment of the present application, a multi-target tracking method is also provided, including: a first determination unit, used to determine a group of to-be-matched trajectories corresponding to an observed target, wherein each to-be-matched trajectory in the group of to-be-matched trajectories is a tracking trajectory of a tracking target of an autonomous driving vehicle, and the observed target is a target observed by the autonomous driving vehicle at a target time; a first matching unit, used to match each to-be-matched trajectory with the observed target according to three-dimensional matching parameters to obtain a three-dimensional matching probability corresponding to each to-be-matched trajectory, wherein the three-dimensional matching parameters are used to evaluate the degree of matching between each to-be-matched trajectory and the observed target in three-dimensional space; a second matching unit, used to match each to-be-matched trajectory with the observed target according to the two-dimensional matching parameters. The method comprises the steps of: matching the trajectories to be matched with the observed target to obtain a two-dimensional matching probability corresponding to each of the trajectories to be matched, wherein the two-dimensional matching parameters are used to evaluate the degree of matching between the two-dimensional projection corresponding to each of the trajectories to be matched and the two-dimensional projection of the observed target on a specified two-dimensional plane; a fusion unit, used to fuse the three-dimensional matching probability corresponding to each of the trajectories to be matched with the two-dimensional matching probability corresponding to each of the trajectories to be matched to obtain a fused matching probability corresponding to each of the trajectories to be matched; and an updating unit, used to update the target tracking trajectory using the observed target when it is determined that the observed target matches the target tracking trajectory in the set of trajectories to be matched based on the fused matching probability corresponding to each of the trajectories to be matched.

[0006] According to another aspect of the embodiments of the present invention, a computer-readable storage medium is provided, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.

[0007] According to another aspect of the embodiments of the present application, a computer program product or a computer program is provided, the computer program product or the computer program includes computer instructions, the computer instructions are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device performs the steps in any of the above method embodiments.

[0008] According to another aspect of an embodiment of the present invention, there is provided an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to execute the steps in any one of the above method embodiments through the computer program.

[0009] Through the embodiments of the present application, a method of adding a projection matching evaluation of the trajectory and the observation is adopted to determine a group of to-be-matched trajectories corresponding to the observation target, wherein each to-be-matched trajectory is a tracking trajectory of a tracking target of the autonomous driving vehicle, and the observation target is a target observed by the autonomous driving vehicle at the target time; each to-be-matched trajectory and the observation target are matched according to the three-dimensional matching parameters to obtain a three-dimensional matching probability corresponding to each to-be-matched trajectory, wherein the three-dimensional matching parameters are used to evaluate the degree of matching between each to-be-matched trajectory and the observation target in three-dimensional space; each to-be-matched trajectory and the observation target are matched according to the two-dimensional matching parameters to obtain a two-dimensional matching probability corresponding to each to-be-matched trajectory, wherein the two-dimensional matching parameters are used to evaluate the two-dimensional projection corresponding to each to-be-matched trajectory and the observation target on a specified two-dimensional plane. the degree of matching between projections; the three-dimensional matching probability corresponding to each trajectory to be matched and the two-dimensional matching probability corresponding to each trajectory to be matched are merged to obtain the fused matching probability corresponding to each trajectory to be matched; when the observed target is determined to match the target tracking trajectory in a set of trajectories to be matched based on the fused matching probability corresponding to each trajectory to be matched, the target tracking trajectory is updated using the observed target. Due to the increase in the projection matching evaluation method between the trajectory and the observation, the projection here is the projection on the specified two-dimensional plane, which can weaken the influence of the three-dimensional distance between the trajectory and the observation on the matching result, so that the trajectory of the same target and the observation can be successfully matched when the three-dimensional distance is far, thereby achieving the technical effect of improving the accuracy of target tracking, and then solving the problem of poor target tracking accuracy in the multi-target tracking method in the related art. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 is a hardware structure block diagram of an optional vehicle-mounted terminal according to an embodiment of the present application;

[0011] Figure 2 is a flowchart of an optional multi-target tracking method according to an embodiment of the present application;

[0012] Figure 3 is a schematic diagram of an optional multi-target tracking method according to an embodiment of the present application;

[0013] Figure 4 is a schematic diagram of another optional multi-target tracking method according to an embodiment of the present application;

[0014] Figure 5 is a flowchart of another optional multi-target tracking method according to an embodiment of the present application;

[0015] Figure 6 is a structural block diagram of an optional multi-target tracking device according to an embodiment of the present application;

[0016] Figure 7 It is a schematic diagram of the structure of an optional electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

[0018] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0019] According to one aspect of an embodiment of the present application, a multi-target tracking method is provided. Optionally, as an optional implementation, Figure 1 FIG. 1 is a hardware structure block diagram of an optional vehicle-mounted terminal according to an embodiment of the present application. Figure 1 As shown, the vehicle terminal may include one or more ( Figure 1 Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the above-mentioned vehicle terminal may also include a transmission device 106 for communication functions and an input and output device 108. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the vehicle-mounted terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0020] The memory 104 can be used to store computer programs, for example, software programs and modules of application software, such as a computer program corresponding to a multi-target tracking method in an embodiment of the present application. The processor 102 executes various functional applications and data processing by running the computer program stored in the memory 104, that is, to implement the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories may be connected to the vehicle-mounted terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0021] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of the vehicle terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, referred to as NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0022] Optionally, the multi-target tracking method in this embodiment can be executed by the vehicle terminal or the background server alone, or can be executed by the vehicle terminal and the background server together. Taking the vehicle terminal as an example, Figure 2 is a flow chart of an optional multi-target tracking method according to an embodiment of the present application, such as Figure 2 As shown, the above method includes:

[0023] Step S202, determining a set of to-be-matched trajectories corresponding to the observed target, wherein each to-be-matched trajectory in the set of to-be-matched trajectories is a tracking trajectory of a tracking target of the autonomous driving vehicle, and the observed target is a target observed by the autonomous driving vehicle at a target time;

[0024] Step S204, matching each to-be-matched trajectory with the observed target according to the three-dimensional matching parameters to obtain a three-dimensional matching probability corresponding to each to-be-matched trajectory, wherein the three-dimensional matching parameters are used to evaluate the matching degree between each to-be-matched trajectory and the observed target in the three-dimensional space;

[0025] Step S206, matching each to-be-matched trajectory with the observed target according to the two-dimensional matching parameters to obtain a two-dimensional matching probability corresponding to each to-be-matched trajectory, wherein the two-dimensional matching parameters are used to evaluate the matching degree between the two-dimensional projection corresponding to each to-be-matched trajectory and the two-dimensional projection of the observed target on a specified two-dimensional plane;

[0026] Step S208, fusing the three-dimensional matching probability corresponding to each trajectory to be matched and the two-dimensional matching probability corresponding to each trajectory to be matched to obtain a fused matching probability corresponding to each trajectory to be matched;

[0027] Step S210 , when it is determined based on the fusion matching probability corresponding to each to-be-matched trajectory that the observed target matches the target tracking trajectory in a set of to-be-matched trajectories, the target tracking trajectory is updated using the observed target.

[0028] The multi-target tracking method in this embodiment can be applied to the field of autonomous driving (unmanned driving), the field of perception fusion, the field of multi-sensor fusion, etc., and can be applied to the scene of multi-target tracking. Taking the field of autonomous driving as an example, multi-target tracking has always been a focus, difficulty and hot spot in the field of autonomous driving. In the related technology, a multi-target tracking algorithm framework is based on the Kalman filter algorithm (Kalman Filter) for trajectory prediction and updating and the Hungarian algorithm (Hungarian algorithm) for correlation matching of trajectory and target observation. This framework is widely used and expanded, and has been optimized and transformed into a variety of multi-target tracking algorithm frameworks. Here, the association matching generally uses the Hungarian algorithm. The Hungarian algorithm comprehensively considers the weights (similarity) between multiple targets and determines the matching relationship between the targets. Therefore, reasonable and accurate weights are one of the prerequisites for high precision of association matching.

[0029] In the actual application of related technologies, if the distance between different sensor observations of the same target is far, they generally cannot be associated and matched together, and thus the observations of the two sensors cannot be fused. For example, when the target is far away from the sensor, the distance detection error of the visual sensor is large, and it is difficult to match the observation with the correct corresponding trajectory through general matching methods, that is, the same target but the observation and trajectory are matched together when the distance is far, thereby reducing the accuracy of multi-target tracking.

[0030] In order to at least partially solve the above technical problems, in this embodiment, when matching the trajectory with the observation, in addition to using the three-dimensional matching parameters to evaluate the observation and the trajectory, a projection matching evaluation method of the trajectory and the observation is added. The projection here is a projection on a specified two-dimensional plane, which can weaken the influence of the three-dimensional distance between the trajectory and the observation on the matching result, so that the trajectory and observation of the same target can be successfully matched when the three-dimensional distance is far, thereby improving the accuracy of target tracking.

[0031] For an autonomous driving vehicle, during its operation, it can observe targets through sensors running on it (e.g., lidar, visual sensors, etc.) to obtain observed targets (e.g., other vehicles, pedestrians, roadblocks, etc.). At each moment, it can observe one or more targets, or no targets. For the observed targets, in order to avoid collision with them, continuously obtain the target status, etc., target tracking can be performed, and trajectories can be established for the tracked targets. Different trajectories can be established for different targets. When a new target is observed, the newly observed target can be matched with the established trajectory to complete the trajectory update, thereby achieving multi-target tracking.

[0032] It should be noted that the algorithm framework steps of multi-target tracking can be divided into: step 1, trajectory prediction; step 2, associating and matching the trajectory with the target observation; step 3, trajectory update; step 4, deleting expired trajectories and establishing new trajectories; step 5, outputting trajectories that meet the output conditions. The accuracy of associating and matching the trajectory with the observation in the above step 2 affects the accuracy of over-target tracking.

[0033] For an observed target (the observed target is a target observed by the autonomous vehicle at the target time), in order to associate and match the trajectory with the observation, a set of to-be-matched trajectories corresponding to the observed target can be determined first; then, each to-be-matched trajectory can be associated and matched with the observed target. Here, each to-be-matched trajectory is a tracking trajectory of a tracking target of the autonomous vehicle. A set of to-be-matched trajectories can be all tracking trajectories established by or for the autonomous vehicle and not expired, or can be a portion of tracking trajectories selected from all tracking trajectories established by or for the autonomous vehicle and not expired. Here, the general multi-target association matching method directly calculates various indicators for all observations and trajectories and directly enters the Hungarian matching algorithm. When there are many targets, the amount of calculation is large and the possibility of mismatching may be increased. However, by removing unreasonable matching pairs between trajectories and observations through some processes, the amount of calculation of subsequent similarity and Hungarian matching can be reduced, and the accuracy of trajectory observation matching can be improved.

[0034] In this embodiment, three-dimensional matching parameters and two-dimensional matching parameters can be used to perform association matching between each track to be matched and the observed target respectively. Here, the three-dimensional matching parameters are used to evaluate the matching degree between each track to be matched and the observed target in three-dimensional space, and the two-dimensional matching parameters are used to evaluate the matching degree between the two-dimensional projection corresponding to each track to be matched and the two-dimensional projection of the observed target on the specified two-dimensional plane. The three-dimensional matching parameters can be considered as three-dimensional evaluation indicators, and the two-dimensional matching parameters can be considered as two-dimensional evaluation indicators. There is a certain angle between the specified two-dimensional plane and the driving direction of the autonomous driving vehicle, and the two are not parallel, otherwise the projection may fail. Optionally, the specified two-dimensional plane can be perpendicular to the horizontal plane, or perpendicular to the driving direction of the autonomous driving vehicle. The specified two-dimensional plane can be a forward camera image plane or parallel to the forward camera image plane, or it can also be a perception plane of other perception devices or parallel to the perception plane of other perception devices. The horizontal plane can be an xy plane of a local coordinate system with a specified position point when the autonomous driving vehicle is started as the coordinate origin, parallel to the driving direction, or it can be a plane parallel to the driving direction with the body coordinate system of the autonomous driving vehicle as a reference.

[0035] Each trajectory to be matched is matched with the observed target according to the three-dimensional matching parameters to obtain the three-dimensional matching probability corresponding to each trajectory to be matched; each trajectory to be matched is matched with the observed target according to the two-dimensional matching parameters to obtain the two-dimensional matching probability corresponding to each trajectory to be matched. The three-dimensional matching parameters can more comprehensively describe the characteristics of the trajectory to be matched and the observed target. These parameters may include but are not limited to: position parameters (three-dimensional coordinates, direction, etc.), motion state parameters (speed, acceleration, etc.) and time parameters (position, posture, etc.). By comprehensively considering these parameters, the similarities and differences between the two can be more accurately evaluated, thereby improving the accuracy of matching. The two-dimensional matching parameters can weaken the influence of the three-dimensional distance on the accuracy of trajectory and observation matching, so that the trajectory and observation of the same target can still be successfully matched when the three-dimensional distance is far.

[0036] For each trajectory to be matched, the corresponding three-dimensional matching probability and the corresponding two-dimensional matching probability can be fused to obtain the corresponding fused matching probability. Here, the three-dimensional matching parameter may include one or more matching parameters, and the two-dimensional matching parameter may include one or more matching parameters. The way to fuse the three-dimensional matching probability and the two-dimensional matching probability may be weighted fusion, and the weights corresponding to different matching parameters may be set based on experience values ​​or test results. Other fusion methods may also be used, which are not limited in this embodiment.

[0037] Based on the fusion matching probability corresponding to each trajectory to be matched, it can be determined whether there is a tracking trajectory that matches the observed target in a group of trajectories to be matched. The above-mentioned determination operation can be performed based on a set matching probability threshold. When there is a trajectory to be matched whose corresponding fusion matching probability is greater than or equal to a specified probability threshold in a group of trajectories to be matched, the trajectory to be matched corresponding to the maximum fusion matching probability is determined as the tracking trajectory that matches the observed target, that is, the target tracking trajectory. In addition, the trajectory to be matched corresponding to the maximum fusion matching probability can also be directly determined as the tracking trajectory that matches the observed target, and the matching of the trajectory and the observation is limited based on the set matching probability threshold (that is, the specified probability threshold), which can avoid mismatching of the trajectory and the observation.

[0038] If there is no corresponding matching track with a fusion matching probability greater than or equal to the specified probability threshold in a set of matching tracks, the observed target can be directly used as a new tracking target to create a tracking track to obtain a tracking track corresponding to the observed target. In order to reduce the targets to be tracked, some observed targets with a shorter appearance time can be filtered out. The target filtering can be performed based on the number of observations, the observed duration, or other parameters, which is not limited in this embodiment.

[0039] Through the above steps, a group of to-be-matched trajectories corresponding to the observed target is determined, wherein each to-be-matched trajectory is a tracking trajectory of a tracking target of the autonomous driving vehicle, and the observed target is a target observed by the autonomous driving vehicle at the target time; each to-be-matched trajectory and the observed target are matched according to the three-dimensional matching parameters to obtain a three-dimensional matching probability corresponding to each to-be-matched trajectory, wherein the three-dimensional matching parameters are used to evaluate the matching degree between each to-be-matched trajectory and the observed target in the three-dimensional space; each to-be-matched trajectory and the observed target are matched according to the two-dimensional matching parameters to obtain a two-dimensional matching probability corresponding to each to-be-matched trajectory, wherein the two-dimensional matching parameters are used to evaluate the matching degree between the two-dimensional projection corresponding to each to-be-matched trajectory and the two-dimensional projection of the observed target on a specified two-dimensional plane; the three-dimensional matching probability corresponding to each to-be-matched trajectory and the two-dimensional matching probability corresponding to each to-be-matched trajectory are fused to obtain a fused matching probability corresponding to each to-be-matched trajectory; when it is determined that the observed target matches the target tracking trajectory in a group of to-be-matched trajectories based on the fused matching probability corresponding to each to-be-matched trajectory, the target tracking trajectory is updated using the observed target, thereby solving the problem of poor target tracking accuracy in the multi-target tracking method in the related art and improving the target tracking accuracy.

[0040] In an exemplary embodiment, determining a set of to-be-matched trajectories corresponding to the observed target includes at least one of the following:

[0041] S11, determining the distance between the predicted position corresponding to each tracking trajectory in the tracking trajectory set and the target position of the observed object, obtaining a reference distance corresponding to each tracking trajectory, and filtering out tracking trajectories whose corresponding reference distances are less than or equal to a first distance threshold from the tracking trajectory set, to obtain an updated tracking trajectory set;

[0042] S12, determining the similarity between the predicted direction corresponding to each tracking trajectory in the tracking trajectory set and the movement direction of the observed target, obtaining a reference similarity corresponding to each tracking trajectory, and filtering out tracking trajectories whose corresponding reference similarities are less than or equal to a specified similarity threshold from the tracking trajectory set, to obtain an updated tracking trajectory set.

[0043] In this embodiment, in order to reduce the number of tracking trajectories that need to be matched during association matching, thereby reducing the calculation amount of association matching and improving the accuracy of association matching, the tracking trajectories of the tracking targets of the autonomous driving vehicle can be screened, that is, some tracking trajectories are screened out from the tracking trajectory set as a set of trajectories to be matched corresponding to the observed targets. Here, the tracking trajectory set is a set of tracking trajectories of the tracking targets of the autonomous driving vehicle. The parameters used for tracking trajectory screening may include but are not limited to at least one of the following: Euclidean distance, cosine similarity of motion direction.

[0044] As an optional implementation, for trajectories and observations with too large a distance difference, the two are not considered to correspond to the same target, and this part of the tracking trajectory can be eliminated. For a set of tracking trajectories, the distance between the predicted position corresponding to each tracking trajectory and the target position of the observed target can be determined to obtain a reference distance corresponding to each tracking trajectory; the tracking trajectories whose corresponding reference distance is less than or equal to the first distance threshold are screened out from the tracking trajectory set to obtain an updated tracking trajectory set. The predicted position corresponding to each tracking trajectory is the predicted position of the tracking target corresponding to each tracking trajectory at the target moment. Here, for each tracking trajectory, trajectory prediction can be performed, that is, the target information of the tracking target corresponding to each tracking trajectory at the target moment is predicted. The predicted target information may include but is not limited to: the predicted position of the tracking target (which can be represented in the form of a prediction box and / or a prediction center point). The first distance threshold may be a distance threshold set based on an empirical value, which can be modified according to actual conditions.

[0045] For example, let w be the final weight, T be a certain trajectory, and O be a certain observation. First, calculate the Euclidean distance between the trajectory and the observation center point coordinates. The coordinates of trajectory T (i.e., the predicted center point coordinates of the tracking target corresponding to the trajectory at the observation time) are (x1, y1), and the center point coordinates of observation 0 are (x2, y2). The Euclidean distance between the two is shown in formula (1):

[0046]

[0047] Then, determine whether the Euclidean distance meets the threshold condition. When the Euclidean distance between the trajectory and the observation is far, it is generally not considered that the two correspond to the same target. Considering that in actual engineering, due to the difference in sensor ranging accuracy, the distance may be far but correspond to the same target. Therefore, based on the Euclidean distance, the trajectory and observation with too large distance difference are mainly eliminated to reduce the calculation amount of the overall process. Let D T is the Euclidean distance threshold (for example, 30 meters, which can be set and modified according to actual conditions). The method of setting w based on the Euclidean distance between the trajectory and the observation center coordinates and the Euclidean distance threshold is as follows:

[0048] ifd>D T w=0

[0049] That is, if the Euclidean distance between the trajectory and the observation center coordinates is greater than the Euclidean distance threshold, w is 0.

[0050] As another optional implementation, when the trajectory differs greatly from the observed direction of motion, the possibility that the trajectory is the same target as the observation is very small. Therefore, if the similarity of the direction of motion is small, this part of the tracking estimate can be screened out and does not enter the subsequent calculation. For the tracking trajectory set, the similarity between the predicted direction corresponding to each tracking trajectory in the tracking trajectory set and the direction of motion of the observed target can be determined to obtain a reference similarity corresponding to each tracking trajectory; the tracking trajectories whose corresponding reference similarity is less than or equal to the specified similarity threshold are screened out from the tracking trajectory set to obtain an updated tracking trajectory set. The predicted direction corresponding to each tracking trajectory is the predicted direction of motion of the tracking target corresponding to each tracking trajectory at the target moment. Here, for each tracking trajectory, the target information of the tracking target corresponding to each tracking trajectory at the target moment can be predicted, and the predicted target information can include but is not limited to: the predicted direction of motion of the tracking target (which can be represented in the form of a velocity vector). The specified similarity threshold can be a similarity threshold set based on an empirical value, which can be modified according to actual conditions.

[0051] Here, for the method of using only the distance parameter to filter out the tracking trajectory, a set of to-be-matched trajectories is a set of updated tracking trajectories obtained after filtering out the tracking trajectories using the distance parameter; for the method of using the motion direction parameter to filter out the tracking trajectories, a set of to-be-matched trajectories is a set of updated tracking trajectories obtained after filtering out the tracking trajectories using the motion direction parameter; for the method of using both the distance parameter and the motion direction parameter to filter out the tracking trajectories, a set of to-be-matched trajectories is a set of updated tracking trajectories obtained after filtering out the tracking trajectories using the distance parameter and the motion direction parameter. The order of using the distance parameter and the motion direction parameter to filter out the tracking trajectories can be pre-specified. For example, the distance parameter can be used to filter out the tracking trajectories first, and then the motion direction parameter can be used to filter out the tracking trajectories. For another example, the motion direction parameter can be used to filter out the tracking trajectories first, and then the distance parameter can be used to filter out the tracking trajectories. This is not limited in the present embodiment.

[0052] For example, let V r =[v tx , v ty ] is the velocity vector of the trajectory, V O =[v ox , v oy ] is the observed velocity vector (if the observation information does not have velocity information, the orientation can be used as the motion direction, for example, camera, lidar, etc. observation). First, the motion direction cosine similarity S is calculated according to formula (2): vel :

[0053]

[0054] Then, determine whether the cosine similarity of the motion direction meets the similarity threshold. When the motion direction of the trajectory differs greatly from that of the observation, it is unlikely that the trajectory and the observation are the same target. Therefore, if the similarity of the motion direction is small, there is no need to enter the subsequent calculation. Let D s is the similarity threshold (for example, 0.1, which can be set and modified according to the actual situation). The method of setting w based on the motion direction cosine similarity and the similarity threshold is as follows:

[0055] if|S vel |<D s w=0

[0056] Here, the trajectory with w=0 is a trajectory that fails the determination, and the subsequent association matching steps can be performed for the trajectory that passes the determination.

[0057] Through this embodiment, by adding a matching pair filtering process based on Euclidean distance and motion trend similarity, the number of tracking trajectories required to be matched during association matching can be reduced, thereby reducing the amount of calculation for subsequent association matching, reducing the probability of mismatching, and improving the accuracy of association matching.

[0058] In an exemplary embodiment, each to-be-matched trajectory and the observed target are matched according to the three-dimensional matching parameters to obtain a three-dimensional matching probability corresponding to each to-be-matched trajectory, including:

[0059] S21, performing at least one of the following first matching operations on each to-be-matched trajectory as the current to-be-matched trajectory to obtain a three-dimensional matching probability corresponding to each to-be-matched trajectory, wherein the tracking target corresponding to the current to-be-matched trajectory is the current tracking target:

[0060] When the three-dimensional matching parameter includes an intersection-and-union ratio, determining an intersection-and-union ratio between a top-view projection frame of a currently tracked target and a top-view projection frame of an observed target, and obtaining an intersection-and-union ratio matching probability corresponding to the current trajectory to be matched;

[0061] When the three-dimensional matching parameters include the Mahalanobis distance, the Mahalanobis distance between the current tracking target and the observed target is determined based on the predicted coordinate vector of the current tracking target, the coordinate vector of the observed target and the specified covariance matrix to obtain the current Mahalanobis distance; when the current Mahalanobis distance is less than or equal to the second distance threshold, the difference between 1 and the ratio of the current Mahalanobis distance to the second distance threshold is determined as the distance matching probability corresponding to the current trajectory to be matched; when the current Mahalanobis distance is greater than the second distance threshold, the specified matching probability is determined as the distance matching probability corresponding to the current trajectory to be matched.

[0062] In this embodiment, for each trajectory to be matched, the first matching operation can be performed on it as the current trajectory to be matched, and the three-dimensional matching probability corresponding to each trajectory to be matched can be obtained. There can be one or more three-dimensional matching parameters. For each three-dimensional matching parameter, the first matching operation corresponding to the three-dimensional matching parameter can be performed on the current trajectory to be matched, and the three-dimensional matching probability corresponding to the three-dimensional matching parameter of each trajectory to be matched can be obtained. Optionally, the three-dimensional matching parameters may include but are not limited to at least one of the following: IOU (Intersection over Union), Mahalanobis distance. Among them, IOU generally measures the degree of overlap between the predicted box (Predicted Bounding Box, the bounding box generated by the model) and the ground truth bounding box in the target detection algorithm, which can intuitively reflect the prediction accuracy of the model. Its value is a ratio between 0 and 1. The higher the value, the higher the overlap between the two boxes, that is, the closer the predicted box is to the ground truth box; Mahalanobis distance is a measurement method for calculating the distance between data points of different dimensions.

[0063] As an optional implementation, when the three-dimensional matching parameters include an intersection-and-union ratio, the first matching operation may include: determining the intersection-and-union ratio between the top-view projection frame of the current tracking target and the top-view projection frame of the observation target, and obtaining the intersection-and-union ratio matching probability corresponding to the current track to be matched. Here, the top-view projection frame of the current tracking target and the top-view projection frame of the observation target may be first determined, and then the intersection area and the union area between the top-view projection frame of the current tracking target and the top-view projection frame of the observation target are calculated, and the result of dividing the intersection area by the union area is determined as the intersection-and-union ratio between the top-view projection frame of the current tracking target and the top-view projection frame of the observation target. The obtained intersection-and-union ratio may be used as the intersection-and-union ratio matching probability corresponding to the current track to be matched, or a regression model may be used to predict the intersection-and-union ratio matching probability corresponding to the current track to be matched under a given IOU value.

[0064] Among them, the overhead projection frame of the current tracking target is the projection frame of the prediction frame of the current tracking target at the target time on the horizontal plane (for the case of uneven roads, it can also be the plane where the autonomous driving vehicle is located), the overhead projection frame of the observed target is the projection frame of the target frame of the observed target on the horizontal plane (for the case of uneven roads, it can also be the plane where the autonomous driving vehicle is located), and the three-dimensional matching probability corresponding to the current trajectory to be matched includes the intersection-and-union ratio matching probability corresponding to the current trajectory to be matched.

[0065] For example, the three-dimensional evaluation index used in the calculation of the three-dimensional evaluation index may include IOU. IOU can also be used in the association matching algorithm to calculate the overlap between the trajectory and the top-view projection frame of the observation, which is used as one of the indicators of whether the trajectory matches the observation. The IOU can be calculated using formula (3):

[0066]

[0067] Among them, Area of ​​intersection represents the intersection area, Area of ​​Union represents the union area, Area of ​​T box represents the area of ​​the top-down projection box of the trajectory, and Area of ​​O box represents the area of ​​the top-down projection box of the observation.

[0068] As another optional implementation, when the three-dimensional matching parameters include the Mahalanobis distance, the first matching operation may include: determining the Mahalanobis distance between the current tracking target and the observed target based on the predicted coordinate vector of the current tracking target, the coordinate vector of the observed target and the specified covariance matrix to obtain the current Mahalanobis distance; when the current Mahalanobis distance is less than or equal to a second distance threshold, determining the difference between 1 and the ratio of the current Mahalanobis distance to the second distance threshold as the distance matching probability corresponding to the current trajectory to be matched; when the current Mahalanobis distance is greater than the second distance threshold, determining the specified matching probability as the distance matching probability corresponding to the current trajectory to be matched.

[0069] Here, the predicted coordinate vector of the current tracking target is the predicted coordinate vector of the current tracking target at the target time. The predicted coordinate vector of the current tracking target and the coordinate vector of the observed target can be first determined, and the current Mahalanobis distance is determined based on the two and a specified covariance matrix. The specified covariance matrix is ​​one of the following: the covariance matrix of the current tracking target at the target time, the covariance matrix of the observed target, and can be selected according to the calculation accuracy of the two when used; then, the size relationship between the current Mahalanobis distance and the second distance threshold is judged. If the current Mahalanobis distance is less than or equal to the second distance threshold, the difference between 1 and the ratio between the current Mahalanobis distance and the second distance threshold is determined as the distance matching probability corresponding to the current trajectory to be matched. Otherwise, the specified matching probability is determined as the distance matching probability corresponding to the current trajectory to be matched. The second distance threshold can be set and adjusted as needed. The specified matching threshold is a smaller value, for example, 0.01, 0.02 or other values, and the three-dimensional matching probability corresponding to the current trajectory to be matched includes the distance matching probability corresponding to the current trajectory to be matched.

[0070] Optionally, in this embodiment, the above matching probability calculation process based on the intersection-over-union ratio and / or the Mahalanobis distance may be performed after the tracking trajectory is screened by the Euclidean distance and / or the motion direction cosine similarity.

[0071] For example, the three-dimensional evaluation index used in the calculation of the three-dimensional evaluation index may include the Mahalanobis distance. is the trajectory point coordinate vector, is the observation point coordinate vector, S is the corresponding covariance matrix (the covariance matrix of the trajectory or the observed covariance matrix can be used, depending on the accuracy of the two calculations). The Mahalanobis distance d between the trajectory and the observation can be calculated according to formula (4): M (T, 0):

[0072]

[0073] Then, the obtained Mahalanobis distance d is determinedM Is it greater than the Mahalanobis distance threshold? The Mahalanobis distance threshold is D m The larger the Mahalanobis distance is, the lower the corresponding matching probability is. The matching probability based on the Mahalanobis distance can be calculated according to formula (5):

[0074]

[0075] Through this embodiment, at least one of the IOU and the Mahalanobis distance is used as a three-dimensional matching parameter to calculate the three-dimensional matching probability, wherein the IOU is calculated based on the top-view projection frame, and the matching probability is calculated based on the Mahalanobis distance in combination with a set distance threshold, which can reduce the complexity of the associated matching calculation and improve the convenience of the associated matching.

[0076] In an exemplary embodiment, the observed target may be a vehicle observed by the autonomous driving vehicle, or may be another traffic participant. In the case where the observed target is a vehicle observed by the autonomous driving vehicle, the target frame of the observed target may include a rear frame. The two-dimensional matching parameter may be a parameter (e.g., IOU) related to the projection of the prediction frame of the tracking target corresponding to each to-be-matched trajectory at the target time on a specified two-dimensional plane and the projection of the target frame of the observed target on a specified two-dimensional plane. Considering that there is a certain deviation between the trajectory and the observation of the same target when the three-dimensional distance is far, the overlapping area between the projection of the prediction frame of the trajectory of the same target and the projection of the observed target frame may be small, and the two-dimensional matching probability determined based on the matching parameters related to the projection area such as IOU is not reasonable. In this embodiment, the two-dimensional matching probability corresponding to each to-be-matched trajectory can be determined based on the positional relationship between the projection of the center point (i.e., the prediction center point) of the prediction frame corresponding to each to-be-matched trajectory at the target time on the specified two-dimensional plane and the projection of at least part of the rear frame of the observed target on the specified two-dimensional plane.

[0077] Correspondingly, each to-be-matched trajectory and the observed target are matched according to the two-dimensional matching parameters to obtain a two-dimensional matching probability corresponding to each to-be-matched trajectory, including:

[0078] S31, performing the following second matching operation on each to-be-matched trajectory as the current to-be-matched trajectory to obtain a two-dimensional matching probability corresponding to each to-be-matched trajectory, wherein the tracking target corresponding to the current to-be-matched trajectory is the current tracking target:

[0079] Determine the projection of the predicted center point of the current tracking target on the specified two-dimensional plane to obtain the current center point projection;

[0080] The current center point projection and the observed vehicle tail frame projection are matched according to the two-dimensional matching parameters to obtain the two-dimensional matching probability corresponding to the current trajectory to be matched.

[0081] For each trajectory to be matched, the second matching operation can be performed on it as the current trajectory to be matched, and the two-dimensional matching probability corresponding to each trajectory to be matched can be obtained. There can be one or more two-dimensional matching parameters. For each two-dimensional matching parameter, the second matching operation corresponding to the two-dimensional matching parameter can be performed on the current trajectory to be matched, and the two-dimensional matching probability corresponding to the two-dimensional matching parameter for each trajectory to be matched can be obtained. Optionally, the two-dimensional matching parameter can be a parameter corresponding to the positional relationship between the projection of the predicted center point of the tracking target corresponding to each trajectory to be matched on the specified two-dimensional plane and the projection of at least part of the rear frame of the observed target on the specified two-dimensional plane.

[0082] In this embodiment, the second matching operation may include: determining the projection of the predicted center point of the current tracking target on the specified two-dimensional plane to obtain the current center point projection, and the current center point projection can be represented by the plane coordinates of the specified two-dimensional plane; matching the current center point projection and the observation vehicle tail frame projection according to the two-dimensional matching parameters to obtain the two-dimensional matching probability corresponding to the current track to be matched. Here, the observation vehicle tail frame projection is the projection of at least part of the vehicle tail frame of the observation target on the specified two-dimensional plane. The two-dimensional matching probability is associated with the relative position between the current center point projection and the observation vehicle tail frame projection. The closer the relative position is, the greater the two-dimensional matching probability is. Conversely, the farther the relative position is, the smaller the two-dimensional matching probability is.

[0083] Through this embodiment, based on the positional relationship between the projection of the predicted center point of the tracking target corresponding to the track to be matched on the two-dimensional plane and the projection of the rear frame of the observed target on the same two-dimensional plane, the two-dimensional matching probability corresponding to the track to be matched is calculated, which can improve the accuracy of the associated matching.

[0084] In an exemplary embodiment, the observation tail frame projection is the projection of the diagonal vertices of the observation target's tail frame on a specified two-dimensional plane. Since the diagonal vertices of the observation target's tail frame can represent the position of the entire observation target, and compared with the method of projecting the entire observation target's tail frame, it can reduce the required amount of calculation while ensuring the accuracy of the association matching, thereby improving the efficiency of the association matching. The observation tail frame projection includes two projection points (i.e., two projection points of the diagonal vertices of the observation target's tail frame), i.e., a first projection point and a second projection point, and the two coordinate axis directions of the specified two-dimensional plane are respectively a first direction and a second direction.

[0085] For example, the center point of the trajectory (i.e., the center point of the prediction box of the tracking target corresponding to the trajectory at the observation time) and the diagonal vertex of the observation vehicle tail box are projected onto the same two-dimensional plane (e.g., the forward camera image plane), and the coordinates of the center point of the trajectory are set as P t =(x t ,y t , zt ), observe the upper left vertex P of the rear frame olt =(x olt ,y olt , z olt ) Observe the lower right vertex P of the rear frame orb =(x orb ,y orb , z orb ), let the projection matrix be TX_W_C (which includes the rotation matrix and the displacement matrix, representing the transformation matrix from the world coordinate system to the two-dimensional plane coordinate system), then the coordinates of the three points transferred to the two-dimensional coordinate system are:

[0086] O t =(u t , v t , 1) = TX_W_C × P t

[0087] O olt =(u olt , v olt , 1) = TX_W_C × P olt

[0088] O orb =(u orb , v orb , 1) = TX_W_C × P orb

[0089] Among them, O t is the projection of the center point of the trajectory, O olt is the projection of the upper left vertex of the rear frame of the observation vehicle, O orb is the projection of the lower right vertex of the rear frame of the observation vehicle, u t 、u olt and u orb is the coordinate value corresponding to a coordinate axis of the two-dimensional plane, v t 、v olt and v orb is the coordinate value corresponding to the other coordinate axis of the two-dimensional plane.

[0090] Correspondingly, the current center point projection and the observed vehicle tail frame projection are matched according to the two-dimensional matching parameters to obtain the two-dimensional matching probability corresponding to the current trajectory to be matched, including:

[0091] S41, when both the first association condition and the second association condition are satisfied, determining the second specified probability as the two-dimensional matching probability corresponding to the current trajectory to be matched;

[0092] S42, when at least one of the first association condition and the second association condition is not satisfied, determining the minimum value of the first matching probability and the second matching probability as the two-dimensional matching probability corresponding to the current trajectory to be matched.

[0093] If the current center point projection is located in a rectangular area with the first projection point and the second projection point as diagonal vertices, that is, the first association condition and the second association condition are both satisfied, it can be considered that the current track to be matched has a high degree of match with the observed target, and the second specified probability can be determined as the two-dimensional matching probability corresponding to the current track to be matched. Here, the first association condition is that the coordinates of the current center point projection in the first direction are between the coordinates of the first projection point in the first direction and the coordinates of the second projection point in the first direction, and the second association condition is that the coordinates of the current center point projection in the second direction are between the coordinates of the first projection point in the first direction and the coordinates of the second projection point in the second direction. The first association condition defines that the current center point projection is located between the first projection point and the second projection point in the first direction, and the second association condition defines that the current center point projection is located between the first projection point and the second projection point in the second direction. The second specified probability is a set probability, and its value can be set and adjusted as needed, for example, 0.8, 0.9, 1, etc. In this embodiment, the specific value of the second specified probability is not limited.

[0094] If at least one of the first association condition and the second association condition is not satisfied, a matching probability can be calculated from the two coordinate axis directions respectively, and the smaller matching probability can be determined as the two-dimensional matching probability corresponding to the current trajectory to be matched. Here, the matching probability calculated from one coordinate axis direction can be determined according to the distance between the two projection points in the coordinate axis direction and the distance between the current center point projection and the center point of the two projection points in the coordinate axis direction. For the first direction, M can be multiplied by the absolute value of the distance between the first projection point and the second projection point in the first direction divided by the absolute value of the distance between the current center point projection and the center point of the first projection point and the second projection point in the first direction to determine the matching probability corresponding to the first direction, that is, the first matching probability. For the second direction, M can be multiplied by the absolute value of the distance between the first projection point and the second projection point in the second direction divided by the absolute value of the distance between the current center point projection and the center point of the first projection point and the second projection point in the second direction to determine the matching probability corresponding to the second direction, that is, the second matching probability. The two-dimensional matching probability corresponding to the current trajectory to be matched is the smaller of the first matching probability and the second matching probability. M is a positive number greater than or equal to 1. For example, M may be 1.5, 2, 3 or other values.

[0095] For example, the evaluation index is calculated by calculating the relative position relationship between the center point of the trajectory and the observation vehicle tail frame on the two-dimensional plane. The position relationship between the projection of the center point of the trajectory and the projection of the diagonal vertex of the observation vehicle tail frame and the matching probability P corresponding to the two-dimensional evaluation index are pro The corresponding relationship is as follows:

[0096] if u t >u olt &&u t orb &&v t >v olt &&v t <v orb P pro =1.0

[0097]

[0098] Here, the diagonal vertices selected are the upper left vertex and the lower right vertex. For the case where the diagonal vertices are selected as the upper right vertex and the lower left vertex, the judgment logic is similar, that is, it is determined that the projections of the center point of the trajectory in the two coordinate axis directions are both located between the two projection points of the diagonal vertices of the observation vehicle tail frame.

[0099] Through this embodiment, the center point of the trajectory and the diagonal vertices of the observation vehicle tail frame are projected onto the same two-dimensional plane, and the two-dimensional matching probability corresponding to the trajectory is determined based on the positional relationship between the projection of the center point of the trajectory and the projection of the diagonal vertices of the observation vehicle tail frame. This can reduce the required amount of calculation while ensuring the accuracy of the associated matching, thereby improving the efficiency of the associated matching.

[0100] In an exemplary embodiment, the matching probability set of the three-dimensional matching probability corresponding to each trajectory to be matched and the two-dimensional matching probability corresponding to each trajectory to be matched includes N matching probabilities corresponding to each trajectory to be matched, where N is a positive integer greater than or equal to 2.

[0101] Among the target association technology weight calculation methods in related technologies, most are weighted averages of IOU algorithms and indicators such as Mahalanobis distance and Euclidean distance. The weights of various indicators are often based on experience and there is no fixed method support (that is, there is no reasonable and universal method to comprehensively consider multiple indicators). They are often not universal. After replacing sensors or related hardware and software configurations, the weights of each indicator need to be repeatedly readjusted. When other indicators are added, they need to be readjusted, and the reusability and scalability are poor.

[0102] To this end, the three-dimensional matching probability corresponding to each trajectory to be matched and the two-dimensional matching probability corresponding to each trajectory to be matched are fused to obtain a fused matching probability corresponding to each trajectory to be matched, including:

[0103] S51, dividing the product of N matching probabilities corresponding to each trajectory to be matched by the sum of the following two products to determine a fusion matching probability corresponding to each trajectory to be matched: the product of the N matching probabilities corresponding to each trajectory to be matched and the product of the N non-matching probabilities corresponding to each trajectory to be matched.

[0104] ​In this embodiment, the DS evidence theory (a complete theory for dealing with uncertainty problems) is used to comprehensively consider various evaluation indicators to perform multi-dimensional evaluation indicator fusion, and the matching probabilities given by different evaluation indicators are fused to give the final Hungarian matching algorithm weights. There is no need to consider the weights of each indicator, and it is easy to expand and add other evaluation indicators. For each trajectory to be matched, the product of the N matching probabilities corresponding to each trajectory to be matched can be determined to obtain the first product corresponding to each trajectory to be matched, and the product of the N non-matching probabilities corresponding to each trajectory to be matched can be determined to obtain the second product corresponding to each trajectory to be matched; the first product is divided by the sum of the first product and the second product to determine the fusion matching probability corresponding to each trajectory to be matched. Here, the N matching probabilities corresponding to each trajectory to be matched and the N non-matching probabilities corresponding to each trajectory to be matched correspond one to one, and each non-matching probability corresponding to each trajectory to be matched is 1 minus a matching probability corresponding to each trajectory to be matched.

[0105] For example, the event types of association matching are relatively simple, only matching (set as event A) or not matching (set as event B). Let the matching probability given by the evaluation index k be P k , then the mismatch probability is 1-P k (k = 1, 2, 3..., the value of k is determined according to the number of evaluation indicators), then the fusion formula of the multi-dimensional evaluation indicators is shown in formula (6):

[0106]

[0107] Among them, m is the matching probability after fusion,

[0108] By transforming formula (6), we can get the following fusion formula:

[0109]

[0110] Through this embodiment, the DS evidence theory is used to comprehensively consider various evaluation indicators to perform multi-dimensional evaluation indicator fusion, which can improve the reusability and expansibility of the evaluation indicator fusion method.

[0111] In an exemplary embodiment, after fusing the three-dimensional matching probability corresponding to each to-be-matched trajectory and the two-dimensional matching probability corresponding to each to-be-matched trajectory, the method further includes:

[0112] S61, when there is a matching trajectory whose corresponding fusion matching probability is greater than or equal to a specified probability threshold in a group of matching trajectories, determining the matching trajectory corresponding to the maximum fusion matching probability as the target tracking trajectory;

[0113] S62, when there is no corresponding matching trajectory with a fusion matching probability greater than or equal to a specified probability threshold in a group of matching trajectories, and the target tracking condition is met, the observed target is taken as a new tracking target to create a tracking trajectory to obtain a tracking trajectory corresponding to the observed target, wherein the target tracking condition includes at least one of the following: the number of consecutive observations is greater than or equal to a specified number threshold, and the time of continuous observation is greater than or equal to a specified time threshold.

[0114] After fusing the three-dimensional matching probability corresponding to each trajectory to be matched and the two-dimensional matching probability corresponding to each trajectory to be matched, based on the fused matching probability corresponding to each trajectory to be matched, a set probability threshold (i.e., a specified probability threshold) can be used to determine whether there is a tracking trajectory that matches the observed target in a group of trajectories to be matched. The specified probability threshold can be set and adjusted according to actual needs, and its value can be 0.8, 0.9 or other values. If there is a corresponding fused matching probability greater than or equal to the specified probability threshold in a group of trajectories to be matched, the fused matching probability corresponding to the maximum fused probability can be determined as the target tracking trajectory.

[0115] For example, Figure 3 As shown in the figure, the specified probability threshold is set to 0.8, and a group of to-be-matched trajectories includes to-be-matched trajectory 1, to-be-matched trajectory 2, and to-be-matched trajectory 3. The fusion matching probability corresponding to to-be-matched trajectory 1 is 0.53, the fusion matching probability corresponding to to-be-matched trajectory 2 is 0.92, and the fusion matching probability corresponding to to-be-matched trajectory 3 is 0.85. The fusion matching probabilities corresponding to to-be-matched trajectory 2 and to-be-matched trajectory 3 are both greater than the specified probability threshold. Therefore, the to-be-matched trajectory 2 corresponding to the maximum fusion probability of 0.92 can be determined as the target tracking trajectory.

[0116] If there is no corresponding track to be matched whose fusion matching probability is greater than or equal to the specified probability threshold in a set of tracks to be matched, the observed target can be directly used as a new tracking target to create a tracking track, and a tracking track corresponding to the observed target can be obtained. Optionally, in order to avoid creating low-value tracking tracks due to false detection, target disappearance after flashing, etc., a target tracking condition can be set, that is, the condition that the target needs to meet to create a tracking track. The target tracking condition can be that the number of consecutive observations is greater than or equal to the specified number threshold, or that the time of continuous observation is greater than or equal to the specified time threshold. The observation can be performed periodically, so the number of consecutive observations greater than or equal to the specified number threshold can be considered as the continuous observation of the target within a specified time. Correspondingly, if there is no corresponding track to be matched whose fusion matching probability is greater than or equal to the specified probability threshold in a set of tracks to be matched, and the target tracking condition is met, the observed target is used as a new tracking target to create a tracking track, and a tracking track corresponding to the observed target is obtained.

[0117] For example, Figure 4 As shown in the figure, the specified probability threshold is set to 0.8, the specified number threshold is set to 5 times, and a group of to-be-matched tracks includes to-be-matched track 1, to-be-matched track 2, and to-be-matched track 3. Among them, the fusion matching probability corresponding to to-be-matched track 1 is 0.53, the fusion matching probability corresponding to to-be-matched track 2 is 0.22, and the fusion matching probability corresponding to to-be-matched track 3 is 0.65, all of which are less than the specified probability threshold. If the number of consecutive observations of the observed target reaches 5 times, the observed target can be used as a new tracking target to create a tracking track.

[0118] Through this embodiment, for a target that fails to match a trajectory, a tracking trajectory is created for it when the number of consecutive observations or the continuous observation time reaches a set threshold, which can reduce the number of created tracking trajectories, thereby reducing the number of tracking trajectories required for association matching, thereby improving the matching rate of association matching.

[0119] The following is an explanation of the multi-target tracking method in the embodiment of the present application in conjunction with an optional example. This optional example provides a solution for multi-target association matching based on the DS evidence theory algorithm and the Hungarian algorithm, such as Figure 5 As shown, the process of the multi-target tracking method in this optional example may include the following steps:

[0120] Step S501, association matching starts;

[0121] Step S502, calculating the Euclidean distance;

[0122] Step S503, determining whether the Euclidean distance is greater than a threshold, if yes, executing step S504, otherwise, executing step S505;

[0123] Step S504, the weight is set to the lowest;

[0124] Step S505, determining whether the motion trend similarity is greater than a threshold, if so, executing step S504, otherwise, executing step S506;

[0125] Step S506, calculation of three-dimensional evaluation indicators (IOU, Mahalanobis distance, etc., because the evaluation indicators are integrated using DS evidence theory, the evaluation indicators are all the probability of matching the trajectory with the observation);

[0126] Step S507, projecting the trajectory center point and the observation detection frame to two-dimensional plane coordinates;

[0127] Step S508, calculating a two-dimensional evaluation index according to the relative position relationship between the center point of the two-dimensional coordinate trajectory and the observation detection frame;

[0128] Step S509, DS evidence theory integrates various evaluation indicators;

[0129] Step S510, performing Hungarian matching based on the trajectory and the observed weight matrix;

[0130] Step S511, the association matching is completed.

[0131] Compared with the general multi-target tracking association matching method, the multi-target tracking method provided in this optional example adds a matching pair filtering process to improve the calculation speed and reduce the probability of false matching; adds trajectory and observation projection matching indicators, increases the matching indicator dimension, and reduces the probability of missed matching; adds DS evidence theory comprehensive evaluation indicators to improve the rationality, convenience and scalability of matching weight calculation.

[0132] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus a necessary general hardware platform, and of course by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes a number of instructions for a terminal device (which can be a mobile phone, computer, server, or network device, etc.) to execute the methods described in each embodiment of the present application.

[0133] According to another aspect of the embodiments of the present application, a multi-target tracking device is also provided, which is used to implement the multi-target tracking method provided in the above-mentioned embodiments, and will not be repeated hereafter. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0134] Figure 6 is a structural block diagram of an optional multi-target tracking device according to an embodiment of the present application, such as Figure 6 As shown, the device comprises:

[0135] A first determining unit 602 is used to determine a set of to-be-matched trajectories corresponding to an observed target, wherein each to-be-matched trajectory in the set of to-be-matched trajectories is a tracking trajectory of a tracking target of the autonomous driving vehicle, and the observed target is a target observed by the autonomous driving vehicle at a target time;

[0136] A first matching unit 604 is used to match each to-be-matched trajectory with the observed target according to the three-dimensional matching parameters to obtain a three-dimensional matching probability corresponding to each to-be-matched trajectory, wherein the three-dimensional matching parameters are used to evaluate the matching degree between each to-be-matched trajectory and the observed target in the three-dimensional space;

[0137] A second matching unit 606 is used to match each to-be-matched trajectory with the observed target according to the two-dimensional matching parameters to obtain a two-dimensional matching probability corresponding to each to-be-matched trajectory, wherein the two-dimensional matching parameters are used to evaluate the matching degree between the two-dimensional projection corresponding to each to-be-matched trajectory and the two-dimensional projection of the observed target on a specified two-dimensional plane;

[0138] A fusion unit 608, configured to fuse the three-dimensional matching probability corresponding to each to-be-matched trajectory and the two-dimensional matching probability corresponding to each to-be-matched trajectory to obtain a fused matching probability corresponding to each to-be-matched trajectory;

[0139] The updating unit 610 is configured to update the target tracking trajectory using the observed target when it is determined that the observed target matches the target tracking trajectory in a set of trajectories to be matched based on the fusion matching probability corresponding to each trajectory to be matched.

[0140] It should be noted that the first determination unit 602 in this embodiment can be used to execute the above step S202, the first matching unit 604 in this embodiment can be used to execute the above step S204, the second matching unit 606 in this embodiment can be used to execute the above step S206, the fusion unit 608 in this embodiment can be used to execute the above step S208, and the update unit 610 in this embodiment can be used to execute the above step S210.

[0141] Through the embodiment of the present application, a group of to-be-matched trajectories corresponding to the observed target is determined, wherein each to-be-matched trajectory is a tracking trajectory of a tracking target of the autonomous driving vehicle, and the observed target is a target observed by the autonomous driving vehicle at the target time; each to-be-matched trajectory and the observed target are matched according to the three-dimensional matching parameters to obtain a three-dimensional matching probability corresponding to each to-be-matched trajectory, wherein the three-dimensional matching parameters are used to evaluate the matching degree between each to-be-matched trajectory and the observed target in the three-dimensional space; each to-be-matched trajectory and the observed target are matched according to the two-dimensional matching parameters to obtain a two-dimensional matching probability corresponding to each to-be-matched trajectory, wherein the two-dimensional matching parameters are used to evaluate the matching degree between the two-dimensional projection corresponding to each to-be-matched trajectory and the two-dimensional projection of the observed target on a specified two-dimensional plane; the three-dimensional matching probability corresponding to each to-be-matched trajectory and the two-dimensional matching probability corresponding to each to-be-matched trajectory are fused to obtain a fused matching probability corresponding to each to-be-matched trajectory; when it is determined that the observed target matches the target tracking trajectory in a group of to-be-matched trajectories based on the fused matching probability corresponding to each to-be-matched trajectory, the target tracking trajectory is updated using the observed target, thereby solving the problem of poor target tracking accuracy in the multi-target tracking method in the related art and improving the target tracking accuracy.

[0142] In an exemplary embodiment, the set of tracking trajectories of the tracking target of the autonomous driving vehicle is a tracking trajectory set. The first determination unit includes at least one of the following: a first execution module, which is used to determine the distance between the predicted position corresponding to each tracking trajectory in the tracking trajectory set and the target position of the observed target, obtain the reference distance corresponding to each tracking trajectory, and filter out the tracking trajectories whose corresponding reference distance is less than or equal to the first distance threshold from the tracking trajectory set to obtain an updated tracking trajectory set, wherein the predicted position corresponding to each tracking trajectory is the predicted position of the tracking target corresponding to each tracking trajectory at the target time; a second execution module, which is used to determine the similarity between the predicted direction corresponding to each tracking trajectory in the tracking trajectory set and the movement direction of the observed target, obtain the reference similarity corresponding to each tracking trajectory, and filter out the tracking trajectories whose corresponding reference similarity is less than or equal to the specified similarity threshold from the tracking trajectory set to obtain an updated tracking trajectory set, wherein the predicted direction corresponding to each tracking trajectory is the predicted movement direction of the tracking target corresponding to each tracking trajectory at the target time; wherein a set of trajectories to be matched includes at least part of the tracking trajectories in the updated tracking trajectory set.

[0143] In an exemplary embodiment, the first matching unit includes: a third execution module, which is used to perform at least one of the following first matching operations on each trajectory to be matched as the current trajectory to be matched, and obtain a three-dimensional matching probability corresponding to each trajectory to be matched, wherein the tracking target corresponding to the current trajectory to be matched is the current tracking target: when the three-dimensional matching parameters include an intersection-and-union ratio, determine the intersection-and-union ratio between the top-view projection frame of the current tracking target and the top-view projection frame of the observation target, and obtain the intersection-and-union ratio matching probability corresponding to the current trajectory to be matched, wherein the top-view projection frame of the current tracking target is a projection frame of the prediction frame of the current tracking target at the target time on the horizontal plane, the top-view projection frame of the observation target is a projection frame of the target frame of the observation target on the horizontal plane, and the three-dimensional matching probability corresponding to the current trajectory to be matched includes the intersection-and-union ratio matching probability corresponding to the current trajectory to be matched; when the three-dimensional matching parameters include the Mahalanobis distance, Based on the predicted coordinate vector of the current tracking target, the coordinate vector of the observed target and the specified covariance matrix, the Mahalanobis distance between the current tracking target and the observed target is determined to obtain the current Mahalanobis distance; when the current Mahalanobis distance is less than or equal to the second distance threshold, the difference between 1 and the ratio between the current Mahalanobis distance and the second distance threshold is determined as the distance matching probability corresponding to the current trajectory to be matched; when the current Mahalanobis distance is greater than the second distance threshold, the specified matching probability is determined as the distance matching probability corresponding to the current trajectory to be matched; wherein the predicted coordinate vector of the current tracking target is the predicted coordinate vector of the current tracking target at the target time, the specified covariance matrix is ​​one of the following: the covariance matrix of the current tracking target at the target time, the covariance matrix of the observed target, and the three-dimensional matching probability corresponding to the current trajectory to be matched includes the distance matching probability corresponding to the current trajectory to be matched.

[0144] In an exemplary embodiment, the observed target is a vehicle observed by the autonomous driving vehicle. The second matching unit includes: a fourth execution module, which is used to perform the following second matching operation on each to-be-matched trajectory as the current to-be-matched trajectory to obtain a two-dimensional matching probability corresponding to each to-be-matched trajectory, wherein the tracking target corresponding to the current to-be-matched trajectory is the current tracking target: determining the projection of the predicted center point of the current tracking target on a specified two-dimensional plane to obtain the current center point projection, wherein the predicted center point of the current tracking target is the predicted center point of the current tracking target at the target time; matching the current center point projection and the observed vehicle tail frame projection according to the two-dimensional matching parameters to obtain a two-dimensional matching probability corresponding to the current to-be-matched trajectory, wherein the observed vehicle tail frame projection is the projection of at least part of the vehicle tail frame of the observed target on the specified two-dimensional plane, and the two-dimensional matching parameters are parameters corresponding to the positional relationship between the projection of the predicted center point of the tracking target corresponding to each to-be-matched trajectory on the specified two-dimensional plane and the projection of the vehicle tail frame of the observed target on the specified two-dimensional plane.

[0145] In an exemplary embodiment, the observed vehicle tail frame projection is the projection of the diagonal vertices of the vehicle tail frame of the observed target on a specified two-dimensional plane; the observed vehicle tail frame projection includes a first projection point and a second projection point, and the two coordinate axis directions of the specified two-dimensional plane are respectively a first direction and a second direction. The fourth execution module includes: a first determination submodule, which is used to determine the second specified probability as the two-dimensional matching probability corresponding to the current trajectory to be matched when both the first association condition and the second association condition are satisfied; and a second determination submodule, which is used to determine the minimum value of the first matching probability and the second matching probability as the two-dimensional matching probability corresponding to the current trajectory to be matched when at least one of the first association condition and the second association condition is not satisfied, wherein the first matching probability is 2 multiplied by the absolute value of the distance between the first projection point and the second projection point in the first direction divided by the absolute value of the distance between the current center point projection and the center point of the first projection point and the second projection point in the first direction, and the second matching probability is 2 multiplied by the absolute value of the distance between the first projection point and the second projection point in the second direction divided by the absolute value of the distance between the current center point projection and the center point of the first projection point and the second projection point in the second direction; wherein the first association condition is that the coordinate of the current center point projection in the first direction is between the coordinate of the first projection point in the first direction and the coordinate of the second projection point in the first direction, and the second association condition is that the coordinate of the current center point projection in the second direction is between the coordinate of the first projection point in the first direction and the coordinate of the second projection point in the second direction.

[0146] In an exemplary embodiment, the matching probability set of the three-dimensional matching probability corresponding to each trajectory to be matched and the two-dimensional matching probability corresponding to each trajectory to be matched includes N matching probabilities corresponding to each trajectory to be matched, wherein N is a positive integer greater than or equal to 2. The fusion unit includes: a determination module, configured to divide the product of the N matching probabilities corresponding to each trajectory to be matched by the sum of the following two products to determine the fused matching probability corresponding to each trajectory to be matched: the product of the N matching probabilities corresponding to each trajectory to be matched and the product of the N non-matching probabilities corresponding to each trajectory to be matched, wherein each non-matching probability corresponding to each trajectory to be matched is 1 minus a result of a matching probability corresponding to each trajectory to be matched, and the N matching probabilities corresponding to each trajectory to be matched and the N non-matching probabilities corresponding to each trajectory to be matched correspond to each other one by one.

[0147] In an exemplary embodiment, the above-mentioned device also includes: a second determination unit, which is used to, after fusing the three-dimensional matching probability corresponding to each to-be-matched trajectory and the two-dimensional matching probability corresponding to each to-be-matched trajectory, determine the to-be-matched trajectory corresponding to the maximum fusion matching probability as the target tracking trajectory when there is a to-be-matched trajectory whose corresponding fusion matching probability is greater than or equal to the specified probability threshold in a group of to-be-matched trajectories; a creation unit, which is used to, when there is no to-be-matched trajectory whose corresponding fusion matching probability is greater than or equal to the specified probability threshold in a group of to-be-matched trajectories and the target tracking condition is met, create a tracking trajectory with the observed target as a new tracking target to obtain a tracking trajectory corresponding to the observed target, wherein the target tracking condition includes at least one of the following: the number of consecutive observations is greater than or equal to the specified number threshold, and the time of consecutive observations is greater than or equal to the specified time threshold.

[0148] It should be noted that the above modules can be implemented by software or hardware. For the latter, it can be implemented in the following ways, but not limited to: the above modules are all located in the same processor; or the above modules are located in different processors in any combination.

[0149] According to another aspect of the embodiments of the present application, a computer-readable storage medium is provided. The computer-readable storage medium includes a stored program, wherein the program executes the steps of any of the above method embodiments when it is run.

[0150] In an exemplary embodiment, the computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, a ROM, a RAM, a mobile hard disk, a magnetic disk, or an optical disk.

[0151] According to another aspect of the embodiments of the present application, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor is configured to execute the steps in any of the above method embodiments through the computer program. In an exemplary embodiment, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0152] For specific examples in this embodiment, reference may be made to the examples described in the above embodiments and exemplary implementation modes, and this embodiment will not be described in detail herein.

[0153] According to another aspect of the embodiment of the present application, a computer program product is also provided, which includes a computer program / instruction, and the computer program / instruction contains a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 709, and / or installed from the removable medium 711. When the computer program is executed by the central processing unit 701, various functions provided by the embodiment of the present application are executed. The above-mentioned serial numbers of the embodiments of the present application are only for description and do not represent the advantages and disadvantages of the embodiments.

[0154] Figure 7 The computer system structure block diagram of the electronic device used to implement the embodiment of the present application is schematically shown. Figure 7 As shown, the computer system 700 includes a CPU (Central Processing Unit) 701, which can perform various appropriate actions and processes according to the program stored in the ROM 702 or the program loaded from the storage part 708 to the RAM 703. Various programs and data required for system operation are also stored in the random access memory 703. The central processing unit 701, the read-only memory 702 and the random access memory 703 are connected to each other through a bus 704. An I / O (Input / Output) interface 705 is also connected to the bus 704.

[0155] The following components are connected to the I / O interface 705: an input section 706 including a keyboard, a mouse, etc.; an output section 707 including a CRT (Cathode Ray Tube), an LCD (Liquid Crystal Display), etc., and a speaker, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, a modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the input / output interface 705 as needed. A removable medium 711, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 710 as needed so that a computer program read therefrom is installed into the storage section 708 as needed.

[0156] In particular, according to an embodiment of the present application, the process described in each method flow chart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program carried on a computer readable medium, and the computer program contains a program code for executing the method shown in the flow chart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 709, and / or installed from the removable medium 711. When the computer program is executed by the central processor 701, various functions defined in the system of the present application are executed.

[0157] It should be noted that Figure 7 The computer system 700 of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0158] Obviously, those skilled in the art should understand that the above modules or steps of the present application can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order from that herein, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present application is not limited to any specific combination of hardware and software.

[0159] The above are only preferred embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the principles of the present application shall be included in the protection scope of the present application.

Claims

1. A multi-target tracking method, characterized in that: include: Determine a set of to-be-matched trajectories corresponding to the observed target, wherein each to-be-matched trajectory in the set of to-be-matched trajectories is a tracking trajectory of a tracking target of the autonomous driving vehicle, and the observed target is a target observed by the autonomous driving vehicle at a target time; Matching each of the to-be-matched trajectories and the observed target according to the three-dimensional matching parameters to obtain a three-dimensional matching probability corresponding to each of the to-be-matched trajectories, wherein the three-dimensional matching parameters are used to evaluate the degree of matching between each of the to-be-matched trajectories and the observed target in the three-dimensional space; Matching each of the to-be-matched trajectories with the observed target according to a two-dimensional matching parameter to obtain a two-dimensional matching probability corresponding to each of the to-be-matched trajectories, wherein the two-dimensional matching parameter is used to evaluate a matching degree between a two-dimensional projection corresponding to each of the to-be-matched trajectories and a two-dimensional projection of the observed target on a specified two-dimensional plane; The three-dimensional matching probability corresponding to each of the to-be-matched trajectories and the two-dimensional matching probability corresponding to each of the to-be-matched trajectories are fused to obtain a fused matching probability corresponding to each of the to-be-matched trajectories; In a case where it is determined that the observed target matches the target tracking track in the set of tracks to be matched based on the fusion matching probability corresponding to each track to be matched, the target tracking track is updated using the observed target.

2. The method according to claim 1, characterized in that The set of tracking trajectories of the tracking targets of the autonomous driving vehicle is a tracking trajectory set; The step of determining a set of to-be-matched trajectories corresponding to the observed target includes at least one of the following: Determine the distance between the predicted position corresponding to each tracking trajectory in the tracking trajectory set and the target position of the observation target, obtain a reference distance corresponding to each tracking trajectory, and filter out the tracking trajectories whose corresponding reference distance is less than or equal to a first distance threshold from the tracking trajectory set, to obtain an updated tracking trajectory set, wherein the predicted position corresponding to each tracking trajectory is the predicted position of the tracking target corresponding to each tracking trajectory at the target time; Determine the similarity between the predicted direction corresponding to each tracking trajectory in the tracking trajectory set and the motion direction of the observed target, obtain a reference similarity corresponding to each tracking trajectory, and filter out tracking trajectories whose corresponding reference similarity is less than or equal to a specified similarity threshold from the tracking trajectory set, to obtain an updated tracking trajectory set, wherein the predicted direction corresponding to each tracking trajectory is the predicted motion direction of the tracking target corresponding to each tracking trajectory at the target moment; The group of to-be-matched trajectories includes at least part of the updated tracking trajectory set.

3. The method according to claim 1, characterized in that The matching each to-be-matched trajectory with the observed target according to the three-dimensional matching parameters to obtain a three-dimensional matching probability corresponding to each to-be-matched trajectory includes: Perform at least one of the following matching operations on each of the to-be-matched trajectories as the current to-be-matched trajectories to obtain a three-dimensional matching probability corresponding to each of the to-be-matched trajectories, wherein the tracking target corresponding to the current to-be-matched trajectories is the current tracking target: When the three-dimensional matching parameter includes an intersection-and-union ratio, determining the intersection-and-union ratio between the top-view projection frame of the current tracking target and the top-view projection frame of the observation target, and obtaining the intersection-and-union ratio matching probability corresponding to the current track to be matched, wherein the top-view projection frame of the current tracking target is a projection frame of the prediction frame of the current tracking target at the target moment on the horizontal plane, the top-view projection frame of the observation target is a projection frame of the target frame of the observation target on the horizontal plane, and the three-dimensional matching probability corresponding to the current track to be matched includes the intersection-and-union ratio matching probability corresponding to the current track to be matched; When the three-dimensional matching parameter includes the Mahalanobis distance, the Mahalanobis distance between the current tracking target and the observed target is determined based on the predicted coordinate vector of the current tracking target, the coordinate vector of the observed target and the specified covariance matrix to obtain the current Mahalanobis distance; when the current Mahalanobis distance is less than or equal to the second distance threshold, the difference between 1 and the ratio between the current Mahalanobis distance and the second distance threshold is determined as the distance matching probability corresponding to the current trajectory to be matched; when the current Mahalanobis distance is greater than the second distance threshold, the specified matching probability is determined as the distance matching probability corresponding to the current trajectory to be matched; wherein the predicted coordinate vector of the current tracking target is the predicted coordinate vector of the current tracking target at the target time, the specified covariance matrix is ​​one of the following: the covariance matrix of the current tracking target at the target time, the covariance matrix of the observed target, and the three-dimensional matching probability corresponding to the current trajectory to be matched includes the distance matching probability corresponding to the current trajectory to be matched.

4. The method according to claim 1, characterized in that: The observation target is a vehicle observed by the autonomous driving vehicle; The step of matching each to-be-matched trajectory with the observed target according to the two-dimensional matching parameters to obtain a two-dimensional matching probability corresponding to each to-be-matched trajectory includes: The following second matching operation is performed on each of the to-be-matched trajectories as the current to-be-matched trajectories to obtain a two-dimensional matching probability corresponding to each of the to-be-matched trajectories, wherein the tracking target corresponding to the current to-be-matched trajectories is the current tracking target: Determine a projection of a predicted center point of the current tracking target on the specified two-dimensional plane to obtain a current center point projection, wherein the predicted center point of the current tracking target is a predicted center point of the current tracking target at the target time; The current center point projection and the observed vehicle tail frame projection are matched according to the two-dimensional matching parameters to obtain a two-dimensional matching probability corresponding to the current trajectory to be matched, wherein the observed vehicle tail frame projection is a projection of at least part of the vehicle tail frame of the observed target on the specified two-dimensional plane, and the two-dimensional matching parameters are parameters corresponding to the positional relationship between the projection of the predicted center point of the tracking target corresponding to each trajectory to be matched on the specified two-dimensional plane and the projection of the vehicle tail frame of the observed target on the specified two-dimensional plane.

5. The method according to claim 4, characterized in that The observation tail frame projection is the projection of the diagonal vertices of the tail frame of the observation target on the specified two-dimensional plane; the observation tail frame projection includes a first projection point and a second projection point, and the two coordinate axis directions of the specified two-dimensional plane are respectively a first direction and a second direction; The matching of the current center point projection and the observation vehicle tail frame projection according to the two-dimensional matching parameters to obtain a two-dimensional matching probability corresponding to the current trajectory to be matched includes: When both the first association condition and the second association condition are satisfied, determining the second specified probability as the two-dimensional matching probability corresponding to the current trajectory to be matched; In the case where at least one of the first association condition and the second association condition is not satisfied, the minimum value of the first matching probability and the second matching probability is determined as the two-dimensional matching probability corresponding to the current trajectory to be matched, wherein the first matching probability is 2 multiplied by the absolute value of the distance between the first projection point and the second projection point in the first direction divided by the absolute value of the distance between the current center point projection and the center point of the first projection point and the second projection point in the first direction, and the second matching probability is 2 multiplied by the absolute value of the distance between the first projection point and the second projection point in the second direction divided by the absolute value of the distance between the current center point projection and the center point of the first projection point and the second projection point in the second direction; Among them, the first association condition is that the coordinates of the current center point projected in the first direction are between the coordinates of the first projection point in the first direction and the coordinates of the second projection point in the first direction, and the second association condition is that the coordinates of the current center point projected in the second direction are between the coordinates of the first projection point in the first direction and the coordinates of the second projection point in the second direction.

6. The method according to any one of claims 1 to 5, characterized in that The matching probability set of the three-dimensional matching probability corresponding to each to-be-matched trajectory and the two-dimensional matching probability corresponding to each to-be-matched trajectory includes N matching probabilities corresponding to each to-be-matched trajectory, where N is a positive integer greater than or equal to 2; The fusing the three-dimensional matching probability corresponding to each to-be-matched trajectory and the two-dimensional matching probability corresponding to each to-be-matched trajectory to obtain a fused matching probability corresponding to each to-be-matched trajectory includes: The fusion matching probability corresponding to each trajectory to be matched is determined by dividing the product of the N matching probabilities corresponding to each trajectory to be matched by the sum of the following two products: the product of the N matching probabilities corresponding to each trajectory to be matched and the product of the N non-matching probabilities corresponding to each trajectory to be matched, wherein each non-matching probability corresponding to each trajectory to be matched is 1 minus a matching probability corresponding to each trajectory to be matched, and the N matching probabilities corresponding to each trajectory to be matched and the N non-matching probabilities corresponding to each trajectory to be matched correspond to each other one by one.

7. The method according to any one of claims 1 to 5, characterized in that After fusing the three-dimensional matching probability corresponding to each to-be-matched trajectory and the two-dimensional matching probability corresponding to each to-be-matched trajectory, the method further includes: In the case that there is a to-be-matched trajectory whose corresponding fusion matching probability is greater than or equal to a specified probability threshold in the group of to-be-matched trajectories, determining the to-be-matched trajectory corresponding to the maximum fusion matching probability as the target tracking trajectory; When there is no corresponding matching trajectory whose fusion matching probability is greater than or equal to a specified probability threshold in the group of matching trajectories, and the target tracking condition is met, the observed target is taken as a new tracking target to create a tracking trajectory to obtain a tracking trajectory corresponding to the observed target, wherein the target tracking condition includes at least one of the following: the number of consecutive observations is greater than or equal to a specified number threshold, and the time of continuous observation is greater than or equal to a specified time threshold.

8. A multi-target tracking device, characterized in that: include: A first determining unit is configured to determine a set of to-be-matched trajectories corresponding to an observed target, wherein each to-be-matched trajectory in the set of to-be-matched trajectories is a tracking trajectory of a tracking target of the autonomous driving vehicle, and the observed target is a target observed by the autonomous driving vehicle at a target time; a first matching unit, configured to match each of the to-be-matched trajectories with the observed target according to a three-dimensional matching parameter to obtain a three-dimensional matching probability corresponding to each of the to-be-matched trajectories, wherein the three-dimensional matching parameter is used to evaluate a degree of matching between each of the to-be-matched trajectories and the observed target in a three-dimensional space; a second matching unit, configured to match each of the to-be-matched trajectories with the observed target according to a two-dimensional matching parameter to obtain a two-dimensional matching probability corresponding to each of the to-be-matched trajectories, wherein the two-dimensional matching parameter is used to evaluate a matching degree between a two-dimensional projection corresponding to each of the to-be-matched trajectories and a two-dimensional projection of the observed target on a specified two-dimensional plane; A fusion unit, used to fuse the three-dimensional matching probability corresponding to each of the to-be-matched trajectories and the two-dimensional matching probability corresponding to each of the to-be-matched trajectories to obtain a fused matching probability corresponding to each of the to-be-matched trajectories; An updating unit is configured to update the target tracking trajectory using the observed target when it is determined that the observed target matches the target tracking trajectory in the set of trajectories to be matched based on the fusion matching probability corresponding to each of the trajectories to be matched.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the computer program implements the steps of any one of the methods of claims 1 to 7 when executed by a processor.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

Images