Multi-target tracking method and device in shielding scene, equipment and storage medium

Abstract

The invention provides a multi-target tracking method and device in a shielding scene, equipment and a storage medium. The method comprises the following steps: acquiring a point cloud cluster of a tracking object before shielding; filtering the point cloud cluster of the tracking object to obtain probability distribution information of a predicted position; processing the shielded point cloud data of each tracking object to obtain a tracking point cloud cluster; and determining a target tracking object corresponding to the tracking point cloud cluster according to the mahalanobis distance between the tracking point cloud cluster and the probability distribution information. By predicting the probability distribution of the position, the possible position of the tracking object can be better described in the sheltered scene, so that the multi-target tracking can be better realized in the sheltered scene based on the mahalanobis distance between the tracking point cloud cluster and the position probability distribution, and the multi-target tracking effect in the sheltered scene is improved.

Description

Technical Field

[0001] The present application relates to the field of target tracking technology, and in particular to a method, device, equipment and storage medium for multi-target tracking in an occlusion scenario. Background Art

[0002] Target tracking refers to the real-time tracking and identification of moving objects in a scene through technologies such as vision and lidar scanning, and has been widely used in various fields.

[0003] However, in many practical application scenarios, for example, in various obstacle competition scenarios of robots, various occlusion obstacles in the scene often affect the collected scene images or scene point cloud data, thereby affecting the object tracking effect. Especially in the multi-target tracking process, when two objects in close positions are re-collected and recognized after being occluded, the recognition accuracy of each tracked object re-determined by the intersection of object detection frames in the related technology is not ideal. Summary of the invention

[0004] Based on this, it is necessary to provide a multi-target tracking method, device, equipment and storage medium in an occlusion scene to address the above-mentioned technical problems, so as to solve the technical problem that the related technology is not ideal in tracking multiple targets in an occlusion scene.

[0005] In a first aspect, the present application provides a multi-target tracking method in an occlusion scene, comprising:

[0006] Obtain the point cloud cluster of each tracked object before occlusion in the target scene;

[0007] Performing filtering processing on the point cloud clusters of each of the tracked objects to obtain probability distribution information of the predicted position corresponding to each of the tracked objects;

[0008] Processing the point cloud data of each of the tracked objects appearing in the target scene after being blocked to obtain a tracking point cloud cluster;

[0009] According to the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position, a target tracking object corresponding to the tracking point cloud cluster is determined from the tracking objects.

[0010] As a feasible embodiment of the present application, filtering the point cloud clusters of each tracked object to obtain probability distribution information of the predicted position corresponding to each tracked object includes:

[0011] For each tracked object, determining Gaussian distribution parameters of the tracked object on a first axis vector and a second axis vector according to a first point cloud cluster and a second point cloud cluster of the tracked object;

[0012] A covariance matrix is ​​generated according to the Gaussian distribution parameters to perform Kalman filtering on the point cloud cluster of the tracked object to obtain probability distribution information of the predicted position corresponding to the tracked object.

[0013] As a feasible embodiment of the present application, determining the Gaussian distribution parameters of the tracked object on the first axis vector and the second axis vector according to the first point cloud cluster and the second point cloud cluster of the tracked object includes:

[0014] Determining motion state parameters of the tracked object in a preset spatial coordinate system according to the first point cloud cluster and the second point cloud cluster of the tracked object;

[0015] The motion state parameters are statistically analyzed to determine Gaussian distribution parameters of the tracked object on a first axis vector and a second axis vector in a preset spatial coordinate system.

[0016] As a feasible embodiment of the present application, generating a covariance matrix according to the Gaussian distribution parameters to perform Kalman filtering on the point cloud cluster of the tracked object to obtain probability distribution information of the predicted position corresponding to the tracked object includes:

[0017] Generate a covariance matrix according to the Gaussian distribution parameters to perform Kalman filtering on the point cloud cluster of the tracked object to obtain first probability distribution information of the predicted position of the tracked object corresponding to the first predicted moment;

[0018] Updating the covariance matrix according to the first probability distribution information to obtain an updated covariance matrix;

[0019] Based on the updated covariance matrix, Kalman filtering is performed on the point cloud cluster of the tracked object to obtain the second probability distribution information of the predicted position of the tracked object corresponding to the second prediction moment, until the probability distribution information of each predicted position of the tracked object when it appears in the target scene after occlusion is determined.

[0020] As a feasible embodiment of the present application, determining the target tracking object corresponding to the tracking point cloud cluster from the tracking objects according to the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position includes:

[0021] Comparing the Mahalanobis distances between the tracking point cloud cluster and the probability distribution information of the predicted positions corresponding to each of the tracking objects;

[0022] When the minimum Mahalanobis distance among the Mahalanobis distances is less than a preset distance threshold, determining the tracking object corresponding to the minimum Mahalanobis distance as the target tracking object corresponding to the tracking point cloud cluster;

[0023] When the minimum Mahalanobis distance among the Mahalanobis distances is greater than a preset distance threshold, the tracking point cloud cluster is determined as a newly appeared tracking object.

[0024] As a feasible embodiment of the present application, determining the target tracking object corresponding to the tracking point cloud cluster from the tracking objects according to the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position includes:

[0025] Performing filtering processing on the point cloud clusters of each of the tracked objects to obtain probability distribution information of the motion state corresponding to each of the tracked objects;

[0026] Determining motion state information corresponding to each of the tracking point cloud clusters;

[0027] According to a first Mahalanobis distance between the motion state information corresponding to the tracking point cloud cluster and the probability distribution information of the motion state corresponding to the tracking object, and a second Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position, a target tracking object corresponding to the tracking point cloud cluster is determined from the tracking objects.

[0028] As a feasible embodiment of the present application, determining the target tracking object corresponding to the tracking point cloud cluster from the tracking objects according to the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position includes:

[0029] Determining the occlusion duration of each of the tracked objects;

[0030] Weighting the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position corresponding to the tracking object according to the weight corresponding to the occlusion duration to obtain a weighted Mahalanobis distance;

[0031] The target tracking object corresponding to the tracking point cloud cluster is determined from the tracking objects according to the weighted Mahalanobis distance.

[0032] In a second aspect, the present application provides a multi-target tracking device in an occlusion scene, comprising:

[0033] An acquisition module, used to acquire a document image to be identified and a portrait base image corresponding to the document image;

[0034] A processing module, used to process the document image and the portrait base image respectively through a trained feature extraction model to obtain a first image feature of the document image and a second image feature of the portrait base image;

[0035] A determination module is used to determine the multi-target tracking result in the occlusion scene of the document image according to the feature similarity between the first image feature and the second image feature.

[0036] In a third aspect, the present application further provides a computer device, the computer device comprising:

[0037] one or more processors;

[0038] Memory; and

[0039] One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the multi-target tracking method in the occlusion scenario provided above.

[0040] In a fourth aspect, the present application also provides a computer-readable storage medium having a computer program stored thereon, and the computer program is loaded by a processor to execute the multi-target tracking method in the occlusion scenario provided above.

[0041] In a fifth aspect, an embodiment of the present application provides a computer program product or a computer program, the computer program product or the computer program including computer instructions, the computer instructions being 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 multi-target tracking method in the occlusion scene provided above.

[0042] The multi-target tracking method in an occluded scene provided by the embodiment of the present application determines the probability distribution information used to describe the possibility of the tracking object at each predicted position by filtering the point cloud clusters of each tracking object in the target scene before occlusion, so as to facilitate the subsequent clustering to obtain the tracking point cloud cluster in the target scene after the occlusion ends, and can determine the target tracking object corresponding to each tracking point cloud cluster from the tracking object by the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position of each tracking object. This is equivalent to the prior art relying on the detection frame of the object, and matching the tracking object based on the intersection and union ratio of the detection frame of the object and the detection frame of the predicted position. Since the position probability distribution can better describe the possible position of the tracking object in the occluded scene than the detection frame of the predicted position, the multi-target tracking based on the Mahalanobis distance between the tracking point cloud cluster and the position probability distribution has a better performance effect in the occluded scene, and improves the tracking effect of multiple targets in the occluded scene. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0044] Figure 1 A schematic diagram of the steps of a multi-target tracking method in an occlusion scenario provided in an embodiment of the present application;

[0045] Figure 2 A schematic diagram of a flow chart of a method for predicting a probability distribution of a position of a tracked object provided in an embodiment of the present application;

[0046] Figure 3 A schematic diagram of the modeling effect of the object motion change in the polar coordinate space coordinate system provided in the embodiment of the present application;

[0047] Figure 4 A schematic flow chart of the steps of performing Kalman filtering based on a dynamically updated covariance matrix according to an embodiment of the present application;

[0048] Figure 5 A schematic diagram of a process flow of determining a target tracking object corresponding to a tracking point cloud cluster based on Mahalanobis distance provided in an embodiment of the present application;

[0049] Figure 6 A schematic diagram of a process flow for implementing association prediction based on a state dimension provided in an embodiment of the present application;

[0050] Figure 7 A schematic flow chart of a step of weighting the Mahalanobis distance based on the occlusion moment to determine a target tracking object provided by an embodiment of the present application;

[0051] Figure 8 A schematic diagram of the structure of a multi-target tracking device in an occlusion scenario provided by an embodiment of the present application;

[0052] Fig. 9 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0053] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions 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 those skilled in the art without creative work are within the scope of protection of this application.

[0054] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, "plurality" means two or more, unless otherwise clearly and specifically defined.

[0055] In the description of the present application, the word "for example" is used to mean "used as an example, illustration or explanation". Any embodiment described as "for example" in the present application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any technician in the field to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes will not be elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in the present application.

[0056] To facilitate understanding of the multi-target tracking method in an occluded scene provided by an embodiment of the present application, the relevant application background of the multi-target tracking method in an occluded scene is first described. Specifically, target tracking refers to collecting data of a target scene through technologies such as vision or laser radar, and determining object information in the scene based on image recognition or point cloud clustering, and then tracking the object information continuously collected in the scene in real time through state prediction, such as the state transfer equation in the Kalman filter. It has been widely used in fields such as intelligent monitoring, intelligent driving, etc. However, in some actual application scenarios where there are occluded scenes, such as various obstacle competition scenes of robots, the occlusion will cause the data of the collected target scene to lose part of the object information, which will affect the tracking effect of the object to a certain extent. Especially in the process of multi-target tracking, when two objects with similar positions are occluded and re-collected and identified, the recognition accuracy of each tracked object determined by the intersection and union ratio of the object detection frame and the predicted position in the related technology is not ideal.

[0057] In order to solve the tracking failure problem that is easy to occur in the above-mentioned occlusion scene, the present application introduces distribution parameters related to the object dynamics to describe the probability distribution of each possible predicted position of the tracked object after being occluded in the process of state prediction, so as to measure the correlation between the point cloud cluster and the tracked object by using the Mahalanobis distance between the probability distribution of the point cloud cluster where the tracked object reappears in the target scene and the tracked object at each possible predicted position, so as to accurately determine the tracked object corresponding to each point cloud cluster, and ensure the target tracking effect in the occlusion scene. Among them, the multi-target tracking method in the occlusion scene generally refers to a computer program arranged in a multi-target tracking device, and the multi-target tracking device generally refers to a processor arranged in a computer device, and the multi-target tracking device in the computer device executes the multi-target tracking method provided by the embodiment of the present application by running the computer program corresponding to the multi-target tracking method, so as to realize multi-target tracking in the occlusion scene.

[0058] For details, please refer to Figure 1 , Figure 1 The schematic diagram of the step flow of the multi-target tracking method in an occlusion scene provided by the embodiment of the present application includes steps S110 to S140:

[0059] S110, obtaining a point cloud cluster of each tracked object before occlusion in the target scene.

[0060] Before the tracked object in the target scene is occluded, similar to the conventional target tracking method, data in the target scene can be collected by vision or laser radar to track the tracked object when it is not occluded. For example, an image acquisition device such as a camera can be used to obtain an image of the target scene, and target detection and recognition can be performed on the image, and then the tracking of each object can be completed by relying on the intersection over union (IoU) of the detection boxes between objects in the images collected at adjacent moments. Of course, considering that the present application is mainly used for multi-target tracking in occluded scenes, especially for tracking and matching the tracked objects before and after occlusion, and visual information is more susceptible to interference from occlusion than laser radar, so in order to further improve the recognition effect of multi-target tracking in occluded scenes, in the embodiment of the present application, it is selected to obtain the point cloud cluster of the tracked object before occlusion by laser radar.

[0061] Specifically, obtaining the point cloud cluster of each tracked object before occlusion in the target scene based on the laser radar usually includes the following steps:

[0062] (1) Initialize the point cloud data acquisition process, including the estimation of the background point cloud in the target scene. Then, manually calibrate the feature points to perform ICP (Iterative Closest Point) registration to obtain the approximate position of each lidar, and then accurately align the pose based on the point to plane algorithm to obtain the precise pose of the lidar;

[0063] (2) Using the three-dimensional grid map (Voxel Map), the background is distinguished by the occupancy of the grid, thereby filtering the background point cloud and obtaining clean point cloud data. Specifically, it includes creating a background grid map for each lidar, and then using the background point cloud to occupy the grid map to obtain a coarser filter map. Then, based on the position of the lidar, the detection ray emitted from the lidar itself is constructed to detect whether each grid can be directly perceived, and occupy all the grids that cannot be perceived in the grid map, which can effectively filter out the scattered noise points of the lidar, and then expand the occupied grid part. The above steps can effectively improve the filtering efficiency of the background point cloud. Due to the use of the grid map, after initialization, the time complexity of a single point query is O(1), and the spatial complexity is only linearly related to the site size and grid density.

[0064] (3) By selecting a suitable point cloud clustering algorithm, the filtered point cloud is clustered to locate all the tracking objects from the point cloud data of the target scene. The clustering algorithm here can be set based on actual needs. For example, in the occlusion scene, it is not required to predetermine the number of clusters and a certain amount of noise points can be tolerated. Therefore, as a common feasible implementation scheme, a density-based clustering method with noise (Density-Based Spatial Clustering of Applications with Noise, DBSCAN) can be selected to complete the clustering of the filtered point cloud data. Of course, in the actual process, considering that the point cloud density decreases with the increase of the laser radar distance, it is difficult to determine the clustering radius and the minimum clustering point of the DBSCAN algorithm. Therefore, as a further optional scheme of the present application, in the process of clustering the point cloud data through DBSCAN, the relevant parameters of DBSCAN can be dynamically updated based on the quadratic inverse relationship between the laser radar point cloud density and the distance to ensure the clustering effect of the point cloud, thereby obtaining the point cloud clusters of each tracking object before occlusion in the target scene.

[0065] Of course, it should be noted that, since the laser radar will collect and cluster the point cloud of the target scene in real time, multiple point cloud clusters at different times can be obtained. Before the tracked object is blocked, the state of these tracked objects is predicted, and then the intersection and union ratio of the object detection frames detected at adjacent moments can be used to effectively determine each tracked object, and identify the corresponding ID (identification information) for distinction, so as to be used for subsequent matching of the tracked object when blocked. Therefore, the point cloud clusters of each tracked object obtained here before occlusion usually include point cloud cluster data of the tracked object at multiple moments before being blocked.

[0066] S120, performing filtering processing on the point cloud clusters of each tracked object to obtain probability distribution information of the predicted position corresponding to each tracked object.

[0067] In the embodiment of the present application, since the point cloud data of the tracked object cannot be observed after the tracked object is blocked, it is also impossible to use the intersection and union ratio of the object detection frame and the object detection frame at the previous moment to achieve tracking and matching of the tracked object. Therefore, it is usually necessary to consider the state prediction of the point cloud cluster of the tracked object to obtain the possible predicted position of the tracked object so as to achieve position matching before and after the tracked object is blocked. For example, as a common feasible implementation scheme, the state transfer equation can be determined by filtering the point cloud cluster of the tracked object, so as to update and calculate the predicted position of the tracked object when it is blocked based on the point cloud cluster data of the tracked object before the blockage, thereby completing the tracking and matching of the point cloud cluster of the tracked object after the blockage.

[0068] However, in the above process, conventional waves are only used to predict the predicted position of the tracked object, while in the technical solution of the present application, the motion change modeling of the tracked object is reasonably constructed to predict the probability distribution information of the predicted position corresponding to the tracked object, that is, it describes the probability distribution of the tracked object appearing in each different position subsequently.

[0069] Specifically, in order to facilitate the understanding of the above content and make the predicted probability distribution of different predicted positions more consistent with the motion state of the tracked object, the embodiment of the present application will further combine the specific motion change modeling to provide an implementation scheme for predicting the probability distribution of the position of the tracked object. For details, please refer to Figure 2 , Figure 2 A schematic flow chart of a step flow for predicting the probability distribution of the position of a tracked object provided in an embodiment of the present application is described in detail as follows.

[0070] The solution provided in the embodiment of the present application needs to perform similar modeling and processing for each tracked object detected before occlusion. Therefore, in order to simplify the description, only a certain tracked object is used as an example for explanation in the embodiment of the present application. For other tracked objects, they can be processed based on similar processing steps. The embodiment of the present application will not be repeated. Specifically, it includes steps S210 to S220:

[0071] S210 , determining Gaussian distribution parameters of the tracked object on a first axis vector and a second axis vector according to the first point cloud cluster and the second point cloud cluster of the tracked object.

[0072] In the embodiment of the present application, combined with the foregoing description, it can be seen that since the tracking of an object before occlusion needs to rely on point cloud cluster data at multiple times, after each tracked object is determined before occlusion, the point cloud clusters of the tracked object at multiple different times can be collected at the same time. Therefore, the collected point cloud clusters of the tracked object at adjacent times can be used as the first point cloud cluster and the second point cloud cluster to determine the distribution parameters of the tracked object on different axis vectors in a specified space, such as the radial axis vector and angular axis vector in a polar coordinate system, or the transverse axis vector and longitudinal axis vector in a Cartesian coordinate system.

[0073] Of course, in order to make the prediction result fit the motion state of the tracked object in space, therefore, when determining the Gaussian distribution parameters of the tracked object on the first axis vector and the second axis vector, the motion state of the tracked object can usually be determined by the first point cloud cluster and the second point cloud cluster of the tracked object, so as to generate Gaussian distribution parameters in the corresponding space based on the motion state information, so as to facilitate the subsequent further prediction of a more accurate probability distribution of the position, that is, determining the Gaussian distribution parameters of the tracked object on the first axis vector and the second axis vector according to the first point cloud cluster and the second point cloud cluster of the tracked object, specifically including:

[0074] Determining motion state parameters of the tracked object in a preset spatial coordinate system according to the first point cloud cluster and the second point cloud cluster of the tracked object;

[0075] The motion state parameters are statistically analyzed to determine Gaussian distribution parameters of the tracked object on a first axis vector and a second axis vector in a preset spatial coordinate system.

[0076] Specifically, to facilitate understanding of the above content, the present application embodiment will be described using the polar coordinate space coordinate system as an example. Figure 3 , Figure 3 This is a schematic diagram of the modeling effect of the object motion change in the polar coordinate space coordinate system provided by the embodiment of the present application. The details are as follows.

[0077] See also Figure 3 , xk-1 is the position information of the tracked object observed at the moment before time k, x k The corresponding position information of the tracked object observed at time k is hat x k Then the corresponding state transfer equation is x k-1 Process the predicted position information of the tracked object at time k. Obviously, Δx (positive in the direction of velocity) and θ (positive in the counterclockwise direction) can reflect the deviation of the prediction result caused by the change of the motion state of the tracked object. If it is assumed that all probability distributions in the model obey the normal distribution, it can be approximately considered that both Δx and θ satisfy the normal distribution with 0 as the expectation, and the two are independent of each other, that is, Δx~N(0,σ1 2 ) and θ~N(0,σ2 2 ). If the best estimation effect is required, statistics of Δx and θ of the tracked object in the real situation should be used to estimate σ1 and σ2. That is, the statistical results of the first point cloud cluster and the second point cloud cluster at multiple adjacent moments are used to estimate σ1 and σ2. Specifically, the 3σ principle can be used to analyze the limit case of Δx and θ for estimation, where the calculation formula of Δx is:

[0078]

[0079] a k is the acceleration, thus, the maximum acceleration a max The estimated σ1 is:

[0080]

[0081] Similarly, for σ2, we can also use the maximum angular acceleration ω max It is estimated that:

[0082]

[0083] Based on the above method, the bivariate normal distribution of the predicted position in polar coordinates can be obtained. Of course, in addition to obtaining the bivariate normal distribution of the predicted position in the radial axis vector Δx and the angular axis vector θ in the polar coordinate space, the corresponding bivariate normal distribution of the predicted position can also be obtained in the Cartesian space coordinate system. Specifically, by further approximation, when the predicted velocity direction of the tracked object is As the positive direction of the x-axis, a right-hand rectangular coordinate system is established. At this time, when θ is small enough, the x′-axis component of the predicted value can be considered and the y′ axis component Independent of each other, that is, the expected value of the predicted value can be determined in this coordinate system Since the polar coordinate space and the aforementioned Cartesian space coordinates have the following related formula:

[0084]

[0085] in, Therefore, combining the bivariate normal distribution of the predicted position in polar coordinates above, we can get:

[0086]

[0087] Since the above is based on the predicted speed direction of the tracking object As the positive direction of the x-axis, the right-handed rectangular coordinate system x′Oy′ is established. Therefore, it can be further converted into the xOy spatial coordinate system under the target scene according to the conversion formula of the spatial coordinate system, so as to obtain the covariance matrix Q of the predicted position k :

[0088]

[0089] in, It is the basis transformation matrix from the x′Oy′ to the xOy coordinate system, which is usually related to the predicted velocity direction and predicted position of the tracked object.

[0090] S220, generating a covariance matrix according to the Gaussian distribution parameters to perform Kalman filtering on the point cloud cluster of the tracked object to obtain probability distribution information of the predicted position corresponding to the tracked object.

[0091] Combined with the above related description, it can be seen that after the Gaussian distribution parameters that can describe the predicted position of the tracked object are generated, the covariance matrix Q of the predicted position can be generated based on it. k , the covariance matrix Q k It can be used for Kalman filter processing of the point cloud cluster of the tracked object to predict the predicted position corresponding to the tracked object in real time. Specifically, the specific update formula of Kalman filter is as follows:

[0092]

[0093] in, and That is, it describes the expected value of the position of the tracking object at time k and time k-1, P k and P k-1 That is, the covariance matrix corresponding to the position of the tracked object at time k and time k-1 describes the distribution of the position of the tracked object, and F k That is, the kinematic parameters of the target moving in a straight line at a uniform speed without any external force when it complies with rigid body kinematics. Specifically:

[0094]

[0095] It can be seen that the covariance matrix Q determined by the above method is k When the target is blocked and observation data cannot be obtained, the Kalman filter can be used to predict the future position of the target based on the existing state. In this case, the inertia of the object will be used to start inertial drift according to the estimated velocity result at the previous moment, that is, the covariance matrix will also increase directionally.

[0096] It can be seen that in the above process, the modeling model used in the embodiment of the present application is different from the common fixed Q k The Kalman filter usage is different in that Q k will be affected by the speed at time k, so that the probability distribution estimation can be more consistent with the actual situation. At the same time, because the parameter estimation model is modeled from the real scene, the estimated probability distribution has the practical significance of being able to fit the actual movement of the tracked object, so that in the subsequent tracking and matching process, there is a comparable meaning based on the Mahalanobis distance. That is to say, in the aforementioned process of generating the covariance matrix according to the Gaussian distribution parameters to perform Kalman filtering on the point cloud cluster of the tracked object and obtain the probability distribution information of the predicted position corresponding to the tracked object, the covariance matrix will be updated in real time and dynamically. For details, please refer to Figure 4 , Figure 4 The schematic diagram of the steps of performing Kalman filtering based on the dynamically updated covariance matrix provided in the embodiment of the present application, specifically, includes steps S410 to S430:

[0097] S410, generating a covariance matrix according to the Gaussian distribution parameters to perform Kalman filtering on the point cloud cluster of the tracked object to obtain first probability distribution information of a predicted position of the tracked object corresponding to a first predicted moment.

[0098] In an embodiment of the present application, due to the length of time the position of the tracked object is obscured, when performing Kalman filtering on the point cloud cluster of the tracked object, Kalman filtering will first be performed based on the covariance matrix generated by the Gaussian distribution parameters to obtain the first probability distribution information of the predicted position of the tracked object at the first prediction moment, that is, the next prediction moment.

[0099] S420: Update the covariance matrix according to the first probability distribution information to obtain an updated covariance matrix.

[0100] Unlike conventional unobstructed scenarios where the covariance matrix can be updated based on the observed position information, in the embodiments of the present application, since the observed position information of the tracked object cannot be obtained when it is occluded, it is necessary to update the covariance matrix based on the probability distribution information of the predicted position to obtain an updated covariance matrix. Since the accumulation of two normal distributions still obeys the normal distribution, in the embodiments of the present application, the covariance matrix describing the probability distribution of the predicted position is updated based on the probability distribution information, and the obtained updated covariance matrix still uses the normal distribution to describe the probability distribution of the predicted position at the next moment.

[0101] S430, performing Kalman filtering on the point cloud cluster of the tracked object based on the updated covariance matrix to obtain second probability distribution information of the predicted position of the tracked object corresponding to the second prediction moment, until the probability distribution information of each predicted position of the tracked object when it appears in the target scene after occlusion is determined.

[0102] In the embodiment of the present application, the point cloud cluster of the tracked object is further processed by Kalman filtering in combination with the aforementioned updated covariance matrix, so as to continue to update the second probability distribution information of the predicted position of the tracked object at the next predicted moment after occlusion, that is, the second predicted moment, and continue to be used for the covariance matrix Q k Dynamic updates are performed until the tracked object reappears in the target scene after being blocked, and the probability distribution information describing the predicted position where the tracked object may appear at this moment is determined for subsequent tracking and matching.

[0103] Of course, in the specific Kalman filter model, in addition to the aforementioned Q k In addition, it usually also includes the covariance matrix R of the observation error k In contrast, since the observation error is relatively fixed and usually comes from the upper observation source, it is believed that the observation error will not change with the change of the target motion state. Therefore, the limit of the observation error can be reasonably estimated by the 3σ principle as the target radius r, that is:

[0104]

[0105] In addition, other common knowledge in the Kalman filtering process will not be elaborated in detail in the embodiments of the present application.

[0106] S130, processing the point cloud data of each of the tracked objects appearing in the target scene after being blocked, to obtain a tracking point cloud cluster.

[0107] In the embodiment of the present application, when each tracking object reappears in the target scene, that is, based on the DSBCAN clustering method provided above, the point cloud data in the target scene can be clustered to generate several new tracking point cloud clusters. At this time, it is necessary to redetermine the tracking object corresponding to each tracking point cloud cluster, or determine the point cloud cluster as a new tracking object. Especially in the multi-target tracking scene, it is necessary to determine the associated mapping relationship of each tracking object before and after occlusion. Among them, the method of processing and clustering the point cloud data of the tracking object appearing in the target scene after occlusion to obtain the tracking point cloud cluster is similar to the method of determining the point cloud cluster of the tracking object before occlusion, and the embodiment of the present application will not be repeated here.

[0108] S140: Determine, from the tracking objects, a target tracking object corresponding to the tracking point cloud cluster according to the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position.

[0109] In the embodiments of the present application, combined with the foregoing relevant descriptions, it can be seen that since the mathematical modeling process is based on modeling in a real scene, and the estimation of each parameter is also determined based on the motion state of the tracked object, the estimated probability distribution has practical significance for fitting the actual motion of the tracked object. Therefore, the target tracking object corresponding to each tracking point cloud cluster can be determined from the tracked object based on the tracking point cloud cluster actually detected after occlusion, that is, the similarity between the observed value and the predicted position, that is, the predicted value.

[0110] It should be noted that due to occlusion, the collected point cloud clusters are discontinuous. Therefore, it is impossible to determine the target tracking object corresponding to each tracking point cloud cluster by calculating the intersection and union ratio of the tracking point cloud cluster and the predicted position through the conventional IoU method. Moreover, in the technical solution of the embodiment of the present application, the predicted position exists in the form of probability distribution information that obeys the normal distribution, that is, it is impossible to use conventional methods such as Euclidean distance to calculate the similarity between the tracking point cloud cluster and the predicted position. Therefore, the embodiment of the present application uses the Mahalanobis distance that can describe the distance between a point and a distribution to achieve the matching of the tracking point cloud cluster and the tracking object. Specifically, taking the spatial position as an example, for a certain tracking point cloud cluster position x0 and the probability distribution information of the predicted position of a certain tracking object For example, the calculation formula of the Mahalanobis distance between the two is as follows:

[0111]

[0112] Among them, the Mahalanobis distance can reflect the probability density of the current object matching the historical object. The essence of comparing the Mahalanobis distance is to perform maximum likelihood estimation. In terms of qualitative, due to the characteristics of the Euclidean distance, that is, on the expected outward extension line, the farther the distance is expected, the larger the Mahalanobis distance is. Under the equal Euclidean distance, the closer the motion state is to the speed direction, the larger the Mahalanobis distance is. Therefore, by analyzing the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of different predicted positions, the target tracking object corresponding to the tracking point cloud cluster can be determined in the tracking object, especially when there are multiple tracking objects with similar positions before occlusion. Due to the superposition of multiple rounds of expected normal distribution, there will be sufficient differences in the probability distribution information of the predicted position. At this time, based on the Mahalanobis distance, the tracking point cloud cluster of each tracking object after occlusion can be more accurately determined. Specifically, based on the comparison of the Mahalanobis distance, it can usually be determined by determining the minimum Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of different predicted positions to determine the target tracking object. For details, please refer to Figure 5 , Figure 5 A schematic flow chart of a step of determining a target tracking object corresponding to a tracking point cloud cluster based on Mahalanobis distance provided in an embodiment of the present application, specifically, comprising steps S510 to S530:

[0113] S510: Compare the Mahalanobis distances between the tracking point cloud cluster and the probability distribution information of the predicted positions corresponding to each of the tracking objects.

[0114] In an embodiment of the present application, for each tracking point cloud cluster, the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position corresponding to each tracked object can be calculated based on the Mahalanobis distance calculation formula provided above, and then these calculated Mahalanobis distances are compared to determine the minimum Mahalanobis distance therein.

[0115] S520: When the minimum Mahalanobis distance among the Mahalanobis distances is less than a preset distance threshold, determine the tracking object corresponding to the minimum Mahalanobis distance as the target tracking object corresponding to the tracking point cloud cluster.

[0116] In an embodiment of the present application, when the minimum Mahalanobis distance is less than a preset distance threshold, that is, under the premise that the probability distribution of the predicted position of a certain tracking point cloud cluster and a certain tracking object is sufficiently similar, it can be considered that the tracking point cloud cluster is sufficiently similar to the tracking object, that is, the tracking object corresponding to the minimum Mahalanobis distance can be determined as the target tracking object corresponding to the tracking point cloud cluster.

[0117] S530: When the minimum Mahalanobis distance among the Mahalanobis distances is greater than a preset distance threshold, determine the tracking point cloud cluster as a newly appeared tracking object.

[0118] In an embodiment of the present application, corresponding to the above, when the minimum Mahalanobis distance is greater than a preset distance threshold, that is, it is considered that the similarity between a certain tracking point cloud cluster and a most likely tracking object still cannot meet the requirements, then the tracking point cloud cluster can be considered as a newly emerged tracking target, and if necessary, it can be combined with other observation information in the subsequent process, for example, combined with visual observation information for further matching and registration, and the embodiments of the present application will not be repeated here.

[0119] Specifically, the above scheme can be described by the following formula:

[0120]

[0121] Among them, T k-1 represents the tracking target at time k-1,

[0122] And X k It represents the cluster at time k, and this step requires the state at the initial time, so it can be considered that

[0123] Of course, in addition to the aforementioned method of measuring the association between each tracking point cloud cluster and the tracking object based only on the Mahalanobis distance between the tracking point cloud cluster in the position dimension and the probability distribution information of the predicted position corresponding to each tracking object, as a further feasible implementation scheme of the present application, the Mahalanobis distance between the tracking point cloud cluster in the motion state dimension and each tracking object can also be further considered to more accurately achieve the matching between the tracking point cloud cluster and the tracking object. For details, please refer to Figure 6 , Figure 6 A schematic flow chart of a step of implementing association prediction based on a state dimension provided in an embodiment of the present application, specifically, comprising steps S610 to S630:

[0124] S610: Filter the point cloud clusters of each tracked object to obtain probability distribution information of the motion state corresponding to each tracked object.

[0125] In the embodiments of the present application, in combination with the above-mentioned related descriptions, it can be known that in the process of performing Kalman filtering on the point cloud cluster of the tracked object to determine the probability distribution information of the predicted position of each tracked object that may appear, while considering the relationship between displacement, velocity and acceleration, and performing reasonable kinematic modeling of the tracked object, the probability distribution information of the motion state corresponding to each tracked object can also be predicted, for example, including velocity direction, velocity magnitude, acceleration, etc. that obey the normal distribution.

[0126] S620: Determine motion state information corresponding to each of the tracking point cloud clusters.

[0127] In the embodiment of the present application, a certain tracking point cloud cluster can only describe the position information at a certain moment. Therefore, under normal circumstances, further clustering is performed on the point cloud data of the target scene after occlusion to further obtain tracking point cloud clusters at multiple moments. The motion state information corresponding to each tracking point cloud cluster can be determined, such as the aforementioned speed direction, speed magnitude, acceleration and other information.

[0128] S630, determining the target tracking object corresponding to the tracking point cloud cluster from the tracking objects based on a first Mahalanobis distance between the motion state information corresponding to the tracking point cloud cluster and the probability distribution information of the motion state corresponding to the tracking object, and a second Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position.

[0129] Compared with directly using the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of different tracking objects at the predicted positions in the position dimension to determine the association relationship between the tracking object and the tracking point cloud cluster, in the embodiment of the present application, by further combining the Mahalanobis distance between the motion state information corresponding to the tracking point cloud cluster and the predicted probability distribution information of the motion state corresponding to the tracking object, the association relationship between the tracking object and the tracking point cloud cluster can be further and more accurately determined. For example, due to the normalization in the Mahalanobis distance, that is, the distance can be compared under the distribution of different parameters, the first Mahalanobis distance between the motion state information corresponding to the tracking point cloud cluster and the probability distribution information of the motion state corresponding to the tracking object, and the second Mahalanobis distance between the motion state information corresponding to the tracking point cloud cluster and the probability distribution information of the predicted position of the tracking object can be appropriately weighted to obtain the weighted Mahalanobis distance, so that the target tracking object corresponding to the tracking point cloud cluster can be more accurately determined from the tracking object, wherein the method for determining the first Mahalanobis distance between the motion state information and the probability distribution information of the motion state corresponding to the tracking object can refer to the aforementioned calculation formula between the predicted position Mahalanobis distances. Alternatively, the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted positions of multiple tracking objects is small enough and close enough, for example, the difference in the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted positions of the first tracking object and the second tracking object is less than a preset threshold. At this time, there may still be a risk of matching failure if only the position is analyzed. Therefore, the target tracking object corresponding to the tracking point cloud cluster can be determined by continuing to consider the relationship between the size of the Mahalanobis distance between the motion state of the tracking point cloud cluster and the probability distribution information of the corresponding predicted motion state of the first tracking object and the second tracking object.

[0130] By combining the aforementioned position dimension with the probability distribution information in the motion state dimension to achieve the matching of the tracking point cloud cluster and the tracking object, the matching accuracy of the tracking object can be further improved, which is especially suitable for the multi-target tracking process in occluded scenes.

[0131] In addition, in addition to the aforementioned implementation scheme for further improving the matching accuracy of the tracked object based on the probability distribution information in the motion state dimension, it is considered that the probability distribution information of the predicted position is obtained based on the superposition of the probability distributions of the predicted positions at multiple moments, that is, as the occlusion moment increases, the error in position prediction will accumulate, that is, under equal Euclidean distance, the tracked object loses tracking, that is, the longer the occlusion time, the greater the Mahalanobis distance can be considered. Therefore, as a further optional implementation scheme of the present application, when determining the target tracking object corresponding to the tracking point cloud cluster from the tracking object based on the Mahalanobis distance between the probability distribution information of the predicted positions of the tracking point cloud cluster and different tracking objects, it is considered that different tracking objects are occluded at different times, that is, the tracking objects may not be occluded at the same time. Therefore, the time when the tracking object is occluded can be further considered to perform appropriate weight optimization on the Mahalanobis distance, thereby further improving the matching accuracy of the tracked object. For details, please refer to. Figure 7 , Figure 7 A schematic flow chart of a step of weighting the Mahalanobis distance based on the occlusion moment to determine the target tracking object provided in an embodiment of the present application, specifically, comprising steps S710 to S720:

[0132] S710: Determine the occlusion duration of each of the tracked objects.

[0133] In an embodiment of the present application, the occlusion duration of the tracking object can be understood as the occlusion duration between the loss of a clustering cluster of a tracking object and the appearance of a new tracking point cloud cluster in the target scene during the clustering process of the point cloud data of the target scene. It can usually be determined based on a timer or based on the period of point cloud data acquisition.

[0134] S720 , weighting the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position corresponding to the tracked object according to the weight corresponding to the occlusion duration to obtain a weighted Mahalanobis distance.

[0135] In the embodiment of the present application, combined with the above-mentioned relevant description, it can be known that since the probability distribution of the predicted position is accumulated, that is, as the occlusion duration increases, the error of the predicted position corresponding to the tracked object will also be accumulated and superimposed. Therefore, the Mahalanobis distance between the probability distribution information of the predicted position corresponding to the tracking point cloud cluster and the tracked object can be weighted based on the weight corresponding to the occlusion duration, wherein the longer the occlusion duration, the higher the corresponding weight coefficient can be set, that is, the occlusion duration can be set to be positively correlated with the weight, so that the weighted Mahalanobis distance between the probability distribution information of the predicted position corresponding to the tracking point cloud cluster and the tracked object is larger.

[0136] S730: Determine a target tracking object corresponding to the tracking point cloud cluster from the tracking objects according to the weighted Mahalanobis distance.

[0137] In the embodiment of the present application, the Mahalanobis distance between the probability distribution information of the predicted position corresponding to the tracking point cloud cluster and the tracking object is weighted by integrating the weight corresponding to the occlusion duration, and the weighted Mahalanobis distance is used to determine the target tracking object corresponding to the tracking point cloud cluster from the tracking object, so that the target tracking object corresponding to the tracking point cloud cluster can be determined more accurately. For example, when the Mahalanobis distance between the probability distribution information of the predicted position corresponding to the tracking point cloud cluster and two tracking objects, such as the third tracking object and the fourth tracking object, is relatively close, at this time, when the occlusion duration of the third tracking object is higher than that of the fourth tracking object, that is, it is believed that the cumulative error between the probability distribution information of the predicted position corresponding to the fourth tracking object and the actual position of the fourth tracking object is lower than the cumulative error of the third tracking object, that is, the credibility of the tracking point cloud cluster as the fourth tracking object is higher. At this time, the Mahalanobis distance is weighted based on the weight corresponding to the occlusion duration, and the weighted Mahalanobis distance between the probability distribution information of the predicted position corresponding to the fourth tracking object of the tracking point cloud cluster is smaller, so that the fourth tracking object will be selected as the target tracking object corresponding to the tracking point cloud cluster.

[0138] Currently, the weighting of the Mahalanobis distance based on the occlusion duration provided above can also be applied to the weighting of the Mahalanobis distance between motion state information, that is, the features between the above embodiments can be appropriately combined based on actual needs, and the embodiments of this application will not elaborate on the implementation scheme after the combination.

[0139] The multi-target tracking method in an occluded scene provided by the embodiment of the present application determines the probability distribution information used to describe the possibility of the tracking object at each predicted position by filtering the point cloud clusters of each tracking object in the target scene before occlusion, so as to facilitate the subsequent clustering to obtain the tracking point cloud cluster in the target scene after the occlusion ends, and can determine the target tracking object corresponding to each tracking point cloud cluster from the tracking object by the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position of each tracking object. This is equivalent to the prior art relying on the detection frame of the object, and matching the tracking object based on the intersection and union ratio of the detection frame of the object and the detection frame of the predicted position. Since the position probability distribution can better describe the possible position of the tracking object in the occluded scene than the detection frame of the predicted position, the multi-target tracking based on the Mahalanobis distance between the tracking point cloud cluster and the position probability distribution has a better performance effect in the occluded scene, and improves the tracking effect of multiple targets in the occluded scene.

[0140] In order to better implement the multi-target tracking method in an occlusion scene provided in the embodiment of the present application, on the basis of the multi-target tracking method in an occlusion scene provided in the embodiment of the present application, the embodiment of the present application also provides a multi-target tracking device in an occlusion scene, such as Figure 8 As shown, the multi-target tracking device 800 in an occlusion scene includes:

[0141] An acquisition module 810 is used to acquire a point cloud cluster of each tracked object before occlusion in a target scene;

[0142] A prediction module 820, configured to filter the point cloud clusters of each tracked object to obtain probability distribution information of a predicted position corresponding to each tracked object;

[0143] A processing module 830 is used to process the point cloud data of each of the tracked objects appearing in the target scene after being blocked, to obtain a tracking point cloud cluster;

[0144] The determination module 840 is used to determine the target tracking object corresponding to the tracking point cloud cluster from the tracking objects according to the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information.

[0145] In some embodiments of the present application, the prediction module 820 is used to determine, for each tracked object, Gaussian distribution parameters of the tracked object on the first axis vector and the second axis vector according to the first point cloud cluster and the second point cloud cluster of the tracked object;

[0146] A covariance matrix is ​​generated according to the Gaussian distribution parameters to perform Kalman filtering on the point cloud cluster of the tracked object to obtain probability distribution information of the predicted position corresponding to the tracked object.

[0147] In some embodiments of the present application, the prediction module 820 is used to determine the motion state parameters of the tracked object in a preset spatial coordinate system according to the first point cloud cluster and the second point cloud cluster of the tracked object;

[0148] The motion state parameters are statistically analyzed to determine Gaussian distribution parameters of the tracked object on a first axis vector and a second axis vector in a preset spatial coordinate system.

[0149] In some embodiments of the present application, the prediction module 820 is used to generate a covariance matrix according to the Gaussian distribution parameters to perform Kalman filtering on the point cloud cluster of the tracked object to obtain first probability distribution information of the predicted position of the tracked object corresponding to the first prediction moment;

[0150] Updating the covariance matrix according to the first probability distribution information to obtain an updated covariance matrix;

[0151] Based on the updated covariance matrix, Kalman filtering is performed on the point cloud cluster of the tracked object to obtain the second probability distribution information of the predicted position of the tracked object corresponding to the second prediction moment, until the probability distribution information of each predicted position of the tracked object when it appears in the target scene after occlusion is determined.

[0152] In some embodiments of the present application, the determination module 840 is used to compare the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position corresponding to each of the tracked objects;

[0153] When the minimum Mahalanobis distance among the Mahalanobis distances is less than a preset distance threshold, determining the tracking object corresponding to the minimum Mahalanobis distance as the target tracking object corresponding to the tracking point cloud cluster;

[0154] When the minimum Mahalanobis distance among the Mahalanobis distances is greater than a preset distance threshold, the tracking point cloud cluster is determined as a newly appeared tracking object.

[0155] In some embodiments of the present application, the determination module 840 is used to perform filtering processing on the point cloud clusters of each tracked object to obtain probability distribution information of the motion state corresponding to each tracked object;

[0156] Determining motion state information corresponding to each of the tracking point cloud clusters;

[0157] According to a first Mahalanobis distance between the motion state information corresponding to the tracking point cloud cluster and the probability distribution information of the motion state corresponding to the tracking object, and a second Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position, a target tracking object corresponding to the tracking point cloud cluster is determined from the tracking objects.

[0158] In some embodiments of the present application, the determination module 840 is used to determine the occlusion duration of each of the tracked objects;

[0159] Weighting the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position corresponding to the tracking object according to the weight corresponding to the occlusion duration to obtain a weighted Mahalanobis distance;

[0160] The target tracking object corresponding to the tracking point cloud cluster is determined from the tracking objects according to the weighted Mahalanobis distance.

[0161] For the specific limitations of the multi-target tracking device in the occlusion scene, please refer to the limitations of the multi-target tracking method in the occlusion scene above, which will not be repeated here. Each module in the above-mentioned multi-target tracking device in the occlusion scene can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0162] In some embodiments of the present application, the multi-target tracking device 800 in an occlusion scene can be implemented in the form of a computer program. The computer program can be implemented in the following manner: Fig. 9 The computer device shown in the figure is run. The memory of the computer device can store various program modules constituting the multi-target tracking device 800 in the occlusion scene, for example, Figure 8 The acquisition module 810, prediction module 820, processing module 830 and determination module 840 are shown. The computer program composed of various program modules enables the processor to execute the steps of the multi-target tracking method in the occlusion scene of each embodiment of the present application described in this specification.

[0163] For example, Fig. 9 The computer device shown can be Figure 8The acquisition module 810 in the multi-target tracking device 800 in the occlusion scene shown executes step S110. The computer device can execute step S120 through the prediction module 820. The computer device can execute step S130 through the processing module 830. The computer device can execute step S140 through the determination module 840. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external computer device through a network connection. When the computer program is executed by the processor, a multi-target tracking method in an occlusion scene is implemented.

[0164] Those skilled in the art will understand that Fig. 9 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0165] In some embodiments of the present application, a computer device is provided, comprising one or more processors; a memory; and one or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the following steps:

[0166] Obtain the point cloud cluster of each tracked object before occlusion in the target scene;

[0167] Performing filtering processing on the point cloud clusters of each of the tracked objects to obtain probability distribution information of the predicted position corresponding to each of the tracked objects;

[0168] Processing the point cloud data of each of the tracked objects appearing in the target scene after being blocked to obtain a tracking point cloud cluster;

[0169] According to the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position, a target tracking object corresponding to the tracking point cloud cluster is determined from the tracking objects.

[0170] In some embodiments of the present application, a computer-readable storage medium is provided, which stores a computer program. The computer program is loaded by a processor, so that the processor performs the following steps:

[0171] Obtain the point cloud cluster of each tracked object before occlusion in the target scene;

[0172] Performing filtering processing on the point cloud clusters of each of the tracked objects to obtain probability distribution information of the predicted position corresponding to each of the tracked objects;

[0173] Processing the point cloud data of each of the tracked objects appearing in the target scene after being blocked to obtain a tracking point cloud cluster;

[0174] According to the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position, a target tracking object corresponding to the tracking point cloud cluster is determined from the tracking objects.

[0175] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Any reference to memory, storage, information library or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory or optical memory, etc. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM).

[0176] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0177] The above is a detailed introduction to the multi-target tracking method, device, computer equipment and storage medium in an occlusion scenario provided by an embodiment of the present application. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea; at the same time, for technical personnel in this field, according to the idea of ​​the present invention, there will be changes in the specific implementation method and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A multi-target tracking method in an occluded scene, characterized in that: include: Obtain the point cloud cluster of each tracked object before occlusion in the target scene; Performing filtering processing on the point cloud clusters of each of the tracked objects to obtain probability distribution information of the predicted position corresponding to each of the tracked objects; Processing the point cloud data of each of the tracked objects appearing in the target scene after being blocked to obtain a tracking point cloud cluster; According to the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position, a target tracking object corresponding to the tracking point cloud cluster is determined from the tracking objects.

2. The method according to claim 1, characterized in that The filtering process is performed on the point cloud clusters of each of the tracked objects to obtain probability distribution information of the predicted position corresponding to each of the tracked objects, including: For each tracked object, determining Gaussian distribution parameters of the tracked object on a first axis vector and a second axis vector according to a first point cloud cluster and a second point cloud cluster of the tracked object; A covariance matrix is ​​generated according to the Gaussian distribution parameters to perform Kalman filtering on the point cloud cluster of the tracked object to obtain probability distribution information of the predicted position corresponding to the tracked object.

3. The method according to claim 2, characterized in that The step of determining Gaussian distribution parameters of the tracked object on a first axis vector and a second axis vector according to the first point cloud cluster and the second point cloud cluster of the tracked object comprises: Determining motion state parameters of the tracked object in a preset spatial coordinate system according to the first point cloud cluster and the second point cloud cluster of the tracked object; The motion state parameters are statistically analyzed to determine Gaussian distribution parameters of the tracked object on a first axis vector and a second axis vector in a preset spatial coordinate system.

4. The method according to claim 2, characterized in that: The generating a covariance matrix according to the Gaussian distribution parameters to perform Kalman filtering on the point cloud cluster of the tracked object to obtain probability distribution information of the predicted position corresponding to the tracked object includes: Generate a covariance matrix according to the Gaussian distribution parameters to perform Kalman filtering on the point cloud cluster of the tracked object to obtain first probability distribution information of the predicted position of the tracked object corresponding to the first predicted moment; Updating the covariance matrix according to the first probability distribution information to obtain an updated covariance matrix; Based on the updated covariance matrix, Kalman filtering is performed on the point cloud cluster of the tracked object to obtain second probability distribution information of the predicted position of the tracked object corresponding to the second prediction moment, until the probability distribution information of each predicted position of the tracked object when it appears in the target scene after occlusion is determined.

5. The method according to claim 1, characterized in that The step of determining the target tracking object corresponding to the tracking point cloud cluster from the tracking objects according to the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position includes: Comparing the Mahalanobis distances between the tracking point cloud cluster and the probability distribution information of the predicted positions corresponding to each of the tracking objects; When the minimum Mahalanobis distance among the Mahalanobis distances is less than a preset distance threshold, determining the tracking object corresponding to the minimum Mahalanobis distance as the target tracking object corresponding to the tracking point cloud cluster; When the minimum Mahalanobis distance among the Mahalanobis distances is greater than a preset distance threshold, the tracking point cloud cluster is determined as a newly appeared tracking object.

6. The method according to claim 1, characterized in that The step of determining the target tracking object corresponding to the tracking point cloud cluster from the tracking objects according to the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position includes: Performing filtering processing on the point cloud clusters of each of the tracked objects to obtain probability distribution information of the motion state corresponding to each of the tracked objects; Determining motion state information corresponding to each of the tracking point cloud clusters; According to a first Mahalanobis distance between the motion state information corresponding to the tracking point cloud cluster and the probability distribution information of the motion state corresponding to the tracking object, and a second Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position, a target tracking object corresponding to the tracking point cloud cluster is determined from the tracking objects.

7. The method according to any one of claims 1 to 6, characterized in that: The step of determining the target tracking object corresponding to the tracking point cloud cluster from the tracking objects according to the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position includes: Determining the occlusion duration of each of the tracked objects; Weighting the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information of the predicted position corresponding to the tracking object according to the weight corresponding to the occlusion duration to obtain a weighted Mahalanobis distance; The target tracking object corresponding to the tracking point cloud cluster is determined from the tracking objects according to the weighted Mahalanobis distance.

8. A multi-target tracking device in an occluded scene, characterized in that: include: An acquisition module is used to acquire point cloud clusters of each tracked object before occlusion in the target scene; A prediction module, used for filtering the point cloud clusters of each tracked object to obtain probability distribution information of the predicted position corresponding to each tracked object; A processing module, used for processing the point cloud data of each of the tracked objects appearing in the target scene after being blocked, to obtain a tracking point cloud cluster; A determination module is used to determine the target tracking object corresponding to the tracking point cloud cluster from the tracking objects according to the Mahalanobis distance between the tracking point cloud cluster and the probability distribution information.

9. A computer device, characterized in that: The computer device comprises: one or more processors; Memory; and One or more applications, wherein the one or more applications are stored in the memory and configured to be executed by the processor to implement the multi-target tracking method in an occlusion scenario according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and the computer program is loaded by a processor to execute the multi-target tracking method in an occlusion scene according to any one of claims 1 to 7.

Images