Target tracking method, target tracking device and computer storage medium
By combining bounding box detection and convex hull detection, the problems of complex target shapes and low tracking accuracy under environmental changes in the prior art are solved, and higher contour description accuracy and environmental adaptability are achieved.
Patent Information
- Application Number
- CN202510034234.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-08
- Publication Date
- 2025-06-10
AI Technical Summary
The existing target tracking methods have significantly reduced the accuracy and applicability of bounding boxes when the target shape is complex and the bounding box is irregular, and it is difficult to accurately track targets in frequent changes.
The target tracking method combining bounding box detection and convex hull detection is adopted. By acquiring laser point clouds, object detection and convex hull detection are performed, the detection target set and convex hull target set are obtained, and the association match is performed, and the contour state of the associated matching target in the target set is updated.
Improve the accuracy of target profile description, and accurately track the target under complex shapes and environmental changes, enhancing the ability to adapt to occlusion and shape changes.
Smart Images
Figure CN120125612A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the technical field, and particularly to a target tracking method, a target tracking device, and a computer storage medium. Background Art
[0002] Currently, many existing solutions use the bounding box method to detect and track targets. These methods use traditional computer vision algorithms (such as YOLO (You Only Look Once, a real-time object detection algorithm based on deep learning), SSD (Single Shot MultiBox Detector), etc.) to generate bounding boxes, and then use technologies such as Kalman filtering or optical flow tracking for target tracking. Although these methods can achieve good results in simple scenarios, in the case of complex target shapes and irregular boundaries, the accuracy and applicability of the bounding boxes are significantly reduced. Summary of the Invention
[0003] To solve the above technical problems, the present application proposes a target tracking method, a target tracking device, and a computer storage medium.
[0004] To solve the above technical problems, the present application proposes a target tracking method, and the target tracking method includes:
[0005] Obtain the laser point cloud at the current moment;
[0006] Perform target detection on the laser point cloud to obtain a set of detected targets;
[0007] Perform convex hull detection on the laser point cloud to obtain a set of convex hull targets;
[0008] Perform target tracking on the set of detected targets to obtain the set of targets at the current moment;
[0009] Associate and match the set of targets at the current moment with the set of convex hull targets to update the contour states of the associated and matched targets in the set of targets.
[0010] Wherein, the performing target tracking on the set of detected targets to obtain the set of targets at the current moment includes:
[0011] Predict the set of predicted targets at the current moment by using the set of targets at the previous moment;
[0012] Associate and match the set of detected targets and the set of predicted targets at the current moment to obtain the detection target indices of the associated and matched targets;
[0013] Update the associated and matched targets according to the detection target indices to obtain the set of targets at the current moment.
[0014] Among them, the step of associating and matching the detection target set and the prediction target set at the current moment to obtain the detection target index of the associated matching target includes:
[0015] Construct a cost matrix using the detection target boxes in the detection target set and the prediction target boxes in the prediction target set;
[0016] Solve the cost matrix through cost minimization assignment;
[0017] Determine the detection target index of the associated matching target according to the solution value of the cost matrix.
[0018] Among them, the step of updating the associated matching target according to the detection target index to obtain the target set at the current moment includes:
[0019] Obtain the prediction target box and prediction covariance of the associated matching target;
[0020] Obtain the observed target box of the prediction target box using the observation matrix;
[0021] Obtain the innovation covariance using the observation matrix, the prediction covariance, and the measurement noise of the associated matching target;
[0022] Obtain the Kalman gain using the innovation covariance, the observation matrix, and the prediction covariance;
[0023] Update the detection target box and detection covariance of the associated matching target using the Kalman gain;
[0024] Obtain the target set at the current moment according to the updated detection target box and detection covariance.
[0025] Among them, the step of associating and matching the target set at the current moment with the convex hull target set includes:
[0026] Based on each detection target box in the target set at the current moment, traverse all convex hull targets in the convex hull target set to obtain the association value between the detection target box and the convex hull target;
[0027] Add the convex hull target index with the association value greater than the preset association threshold to the association set of the associated matching target.
[0028] Among them, the step of updating the contour state of the associated matching target in the target set includes:
[0029] Merge and cluster the convex hull target point clouds in the association set of the associated matching target, and construct the target contour measurement of the associated matching target;
[0030] Based on the target contour state at the previous moment, obtain the predicted contour state at the current moment;
[0031] Utilize the position information of the detection target box of the associated matching target and the predicted contour state to obtain the predicted measurement of the associated matching target;
[0032] Based on the target covariance at the previous moment, obtain the predicted covariance at the current moment;
[0033] Utilize the predicted measurement and the target contour measurement to generate an innovation matrix;
[0034] Utilize the innovation matrix and the Kalman gain to update the predicted contour state, and obtain the contour state of the associated matching target;
[0035] Utilize the Kalman gain and the observation matrix to update the predicted covariance, and obtain the covariance of the associated matching target.
[0036] Wherein, the target tracking method further includes:
[0037] Based on the predicted covariance and the equivalent measurement noise, construct an innovation covariance;
[0038] Based on the predicted covariance, the innovation covariance and the observation matrix, construct the Kalman gain.
[0039] Wherein, the step of merging and clustering the convex hull target point clouds in the association set of the associated matching target and constructing the target contour measurement of the associated matching target includes:
[0040] Merge and cluster the convex hull target point clouds in the association set of the associated matching target to obtain a clustering result;
[0041] According to the clustering result and the orientation angle of the detection target box, construct the target contour measurement of the associated matching target.
[0042] To solve the above technical problems, the present application also proposes a target tracking device, which includes a memory and a processor coupled to the memory; wherein, the memory is used to store program data, and the processor is used to execute the program data to implement the target tracking method as described above.
[0043] To solve the above technical problems, the present application also provides a computer storage medium for storing program data, which, when executed by a computer, is used to implement the above-mentioned target tracking method.
[0044] Compared with the prior art, the beneficial effects of the present application are as follows: The target tracking device acquires the laser point cloud at the current moment; performs target detection on the laser point cloud to obtain a set of detected targets; performs convex hull detection on the laser point cloud to obtain a set of convex hull targets; performs target tracking on the set of detected targets to obtain the set of targets at the current moment; and associates and matches the set of targets at the current moment with the set of convex hull targets to update the contour states of the associated and matched targets in the set of targets. Through the above target tracking method, the target bounding box results are combined with the convex hull technology set. The bounding box is used for preliminary tracking, and the true contour of the target is extracted through the convex hull, thereby improving the accuracy of contour description. Description of the Drawings
[0045] To more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.
[0046] Among them:
[0047] Figure 1 is a schematic flowchart of an embodiment of the target tracking method provided by the present application;
[0048] Figure 2 is a schematic overall flowchart of the target tracking method provided by the present application;
[0049] Figure 3 is Figure 1 a specific schematic flowchart of step S14 of the target tracking method shown;
[0050] Figure 4 is Figure 1 a specific schematic flowchart of step S15 of the target tracking method shown;
[0051] Figure 5 is a schematic structural diagram of an embodiment of the target tracking device provided by the present application;
[0052] Figure 6 is a schematic structural diagram of an embodiment of the target tracking device provided by the present application;
[0053] Figure 7 is a schematic structural diagram of an embodiment of the computer storage medium provided by the present application. Detailed Embodiments
[0054] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present application without creative efforts shall fall within the protection scope of the present application.
[0055] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned accompanying drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present application described herein, for example, can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.
[0056] Current target tracking solutions use point cloud clustering techniques (such as DBSCAN (Density-Based Spatial Clustering of Applications with Noise), Euclidean clustering, etc.) to identify targets. These methods form targets by clustering adjacent points, but in a dynamic environment, the clustering results may be affected by noise and occlusion, resulting in inaccurate identification.
[0057] With the development of deep learning technology, more and more research has begun to use convolutional neural networks (CNNs, Convolutional Neural Networks) and point cloud processing networks (such as PointNet, PointRCNN, etc.) for target detection and tracking. These deep learning methods perform well in feature extraction, but usually require a large amount of labeled data for training and consume a large amount of computing resources. At the same time, deep learning methods may still be unable to effectively maintain an accurate description of the target shape when the target is occluded or partially visible.
[0058] Therefore, the technical problems solved by the target tracking method of the present application are as follows:
[0059] 1. The point cloud object tracking algorithm based on the target bounding box uses the minimum circumscribed cuboid of the target point cloud as the contour of the target for tracking. The disadvantage of this method is that it cannot accurately capture the actual contour of the target. Especially when the target shape is complex, the tracking accuracy will be affected.
[0060] 2. In an environment with frequent changes, the movement of the target and external interferences (such as occlusion and noise, etc.) may cause fluctuations in the target bounding box, resulting in fluctuations in the tracking results.
[0061] 3. During the tracking process of the target, existing methods often lack an effective memory of the accurate historical contour of the target and cannot describe the occluded contour of the target based on historical information.
[0062] To address the above disadvantages, the present application aims to improve the accuracy of target contour tracking. By combining bounding box extraction and convex hull extraction, the limitations of traditional methods are overcome. In addition, the present application also adopts a method of tracking the target contour, so that even in the case of partial occlusion, the contour of the target can be reasonably inferred, thereby improving the overall accuracy and consistency of tracking.
[0063] For details, please continue to refer to Figure 1 and Figure 2 , Figure 1 which is a schematic flowchart of an embodiment of the target tracking method provided by the present application, Figure 2 and
[0064] The target tracking method of the present application is applied to a target tracking device. Among them, the target tracking device of the present application can be a server, or a terminal device, or a system in which the server and the terminal device cooperate with each other. Correspondingly, each part included in the target tracking device, such as each unit, sub-unit, module, and sub-module, can be all set in the server, or all set in the terminal device, or can be respectively set in the server and the terminal device.
[0065] Furthermore, the above server can be hardware or software. When the server is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or as a single server. When the server is software, it can be implemented as multiple software or software modules, such as software or software modules used to provide a distributed server, or can be implemented as a single software or software module, which is not specifically limited herein.
[0066] As Figure 1 shown, the specific steps are as follows:
[0067] Step S11: Obtain the laser point cloud at the current moment.
[0068] In the embodiment of the present application, the target tracking device acquires the point cloud obtained by the lidar at time t where N t is the number of point clouds at time t, The point cloud is a vector composed of the x, y, and z three-axis coordinates and the intensity information i in the Cartesian coordinate system.
[0069] Step S12: Perform target detection on the laser point cloud to obtain a set of detected targets.
[0070] In the embodiment of the present application, taking the Kalman filter as an example for the tracking algorithm based on the target bounding box, the target tracking device sends the point cloud into the target detection module, and thus obtains a set of detected targets where M t is the number of detected targets at time t, and the detection information of the target is composed of the three-axis coordinates of the target center point x, y, z, the length, width, and height l, w, h of the target box, and the orientation angle θ.
[0071] Step S13: Perform convex hull detection on the laser point cloud to obtain a set of convex hull targets.
[0072] In the embodiment of the present application, taking the Gaussian process as an example for the target contour tracking algorithm based on the convex hull, the target tracking device sends the point cloud into the convex hull detection module to obtain a set of convex hull targets where L t is the number of detected convex hulls at time t, and each convex hull is described by a series of point clouds.
[0073] Step S14: Perform target tracking on the set of detected targets to obtain the set of targets at the current moment.
[0074] In the embodiment of the present application, the target tracking device uses K t to represent the number of targets tracked at time t, and uses to represent the target state, which includes the positions and velocities of the x, y, and z three axes of the target center point, the length, width, and height l, w, h of the target box, and the orientation angle θ.
[0075] Specifically, the present application provides a specific target tracking solution. Please continue to refer to Figure 3 , Figure 3 is Figure 1 the specific process schematic diagram of step S14 of the target tracking method shown.
[0076] As Figure 3 shown, the specific steps are as follows:
[0077] Step S141: Predict the set of predicted targets at the current moment using the set of targets at the previous moment.
[0078] In the embodiment of the present application, the target tracking device uses the target state at the previous moment and covariance to predict the K t targets at time t:
[0079]
[0080] where F is the state transition matrix, is the process noise matrix.
[0081] Step S142: Correlate and match the detection target set and the predicted target set at the current moment to obtain the detection target index of the correlated and matched target.
[0082] In the embodiment of the present application, the target tracking device uses the Hungarian algorithm to perform correlation matching on the detection target set Z t and the predicted target set for correlation matching.
[0083] The target tracking device constructs a cost matrix C using the IoU of the target bounding box and the Euclidean distance of the center point between the predicted target and the detection target t :
[0084]
[0085] The calculation formula for each element in this matrix is as follows:
[0086]
[0087] where α and β are normalization parameters used to control the weights between IoU and the Euclidean distance, and ε is a small constant used to avoid the denominator being zero.
[0088] Among them, the target tracking device uses the Hungarian algorithm to achieve minimum cost allocation, and records the allocation result as where, represents the detection target index associated with target k.
[0089] Step S143: Update the correlated and matched target according to the detection target index to obtain the target set at the current moment.
[0090] In the embodiment of the present application, for each target k, the target tracking device calculates the predicted measurement innovation covariance and Kalman gain using the observation matrix H and the measurement noise R. The specific formulas are as follows:
[0091]
[0092] Then, the target tracking device uses the associated detected target to perform an update:
[0093]
[0094] Thereby, the estimation of the target state and covariance at time t is achieved, and finally the target set at the current time is obtained
[0095] Step S15: Associate and match the target set at the current time with the convex hull target set to update the contour state of the associated and matched targets in the target set.
[0096] In the embodiment of the present application, the target tracking device uses the updated target set χ t and the convex hull target set to achieve the tracking of the convex hull. Specifically, the convex hull target is used to construct and update the contour state of the target, that is, to describe the actual occupied part of the target in the point cloud.
[0097] Specifically, the present application provides a specific contour state update scheme. Please continue to refer to Figure 4 , Figure 4 is Figure 1 the specific flow schematic diagram of step S15 of the target tracking method shown.
[0098] As Figure 4 shown, the specific steps are as follows:
[0099] Step S151: Based on each detected target box in the target set at the current time, traverse all convex hull targets in the convex hull target set to obtain the association value between the detected target box and the convex hull target.
[0100] In the embodiment of the present application, the target tracking device models the target contour state as a star convex shape and uses to describe the contour state of target k. Where represents the contour point coordinates of the target in polar coordinates with the target center point as the origin. Where θ i uses to represent the angle, and N points in [0, 2π) are evenly sampled on, that is, θ i = 2π(i - 1) / N, and the radial function f k (θ i ) represents the contour radius corresponding to each angle, and a Gaussian process is used for modeling:
[0101]
[0102] Among them, κ f (θ i , θ i') is the covariance function, and the squared exponential kernel is generally selected according to this scenario:
[0103]
[0104] In the above modeling, σ r is the standard deviation of the radius, σ f is the standard deviation of the radial function, and l is the scale factor, all of which are hyperparameters of the Gaussian process.
[0105] Furthermore, the target tracking device extracts the target coordinates of the center point of the target box and the information of the length, width, and height, and performs cross-matching with the convex hull target point cloud for cross-matching.
[0106] For each target k, the target tracking device traverses the convex hull targets l ∈ {1,..., L t}, and calculates the ratio r k,l of the overlapping area between the convex hull target and the target to its own area:
[0107]
[0108] where inter() is the function for calculating the overlapping area, and area() is the function for calculating the convex hull area.
[0109] Step S152: Add the convex hull target index with an association value greater than the preset association threshold to the association set of the associated matching target.
[0110] In the embodiment of the present application, if r k,l > t k , where t k is the association threshold of target k, then add the convex hull target index l to the association set of target k.
[0111] It should be noted that the association threshold of the present application can be set separately for different target categories.
[0112] Step S153: Merge and cluster the convex hull target point clouds in the association set of the associated matching target, and construct the target contour measurement of the associated matching target.
[0113] In the embodiment of the present application, the target tracking device merges the convex hull target point clouds in the association set of target k to construct the convex hull point cloud associated with target k Then cluster the point cloud, and use the interpolation method to construct the target contour measurement according to the angle θ i corresponding to the target contour state
[0114] Specifically, the target tracking device constructs a measurement model.
[0115] Assume that at time t, target k generates M k measurements. For the j-th measurement where j ∈ {1, …, M k}, the following measurement model can be constructed:
[0116]
[0117] where the measurement noise w k,j ~N(0, R k,j ), and the ideal measurement is calculated as follows:
[0118]
[0119] where pos() is a function for extracting the position, used to extract the position information in the target state, and p k,j is the rotation factor:
[0120]
[0121] where ||·|| 2 is the Euclidean norm, is the angle of the current measurement in the local coordinate system of the target, is the angle of the measurement in the global coordinate system, yaw() is a function for extracting the orientation angle, represents the radial distance between the measurement and the center point of the target.
[0122] The target tracking device uses a Gaussian process to model the radial function as follows:
[0123]
[0124] where,
[0125] w f,k,j ~N(0, R f,k,j )
[0126]
[0127] where, is the covariance matrix, and its elements can be expressed as follows:
[0128]
[0129] In summary, the measurement model can be expressed as follows:
[0130]
[0131] where, is the equivalent measurement noise:
[0132]
[0133] Step S154: Obtain the predicted contour state at the current moment based on the target contour state at the previous moment.
[0134] In the embodiment of the present application, the target tracking device predicts the contour state according to the estimation of the target contour state at the previous moment:
[0135]
[0136] where is the target contour state transition matrix, is the process noise.
[0137] Step S155: Use the position information of the detected target box of the associated matching target and the predicted contour state to obtain the predicted measurement of the associated matching target.
[0138] In the embodiment of the present application, the target tracking device regards the edge points of the convex hull as the measurements corresponding to the target contour, and updates the target contour state accordingly. For the sake of simplicity, the time index is omitted in this step.
[0139] For each angle, the corresponding predicted measurement can be calculated as follows:
[0140]
[0141] Step S156: Obtain the predicted covariance at the current moment based on the target covariance at the previous moment.
[0142] Step S157: Generate an innovation matrix using the predicted measurement and the target contour measurement.
[0143] In the embodiment of the present application, the innovation covariance is as follows:
[0144]
[0145] Thus, the Kalman gain can be constructed:
[0146]
[0147] Furthermore, the target tracking device updates the target contour state using the measurements generated in the above process, and generates an innovation matrix using the predicted measurement and the target measurement:
[0148]
[0149] Step S158: Update the predicted contour state using the innovation matrix and the Kalman gain to obtain the contour state of the associated matching target.
[0150] Step S159: Update the predicted covariance using the Kalman gain and the observation matrix to obtain the covariance of the associated matching target.
[0151] In the embodiment of the present application, the target tracking device updates the target contour state and covariance as follows:
[0152]
[0153] In the present application, the target tracking device acquires the laser point cloud at the current moment; performs target detection on the laser point cloud to obtain a set of detected targets; performs convex hull detection on the laser point cloud to obtain a set of convex hull targets; performs target tracking on the set of detected targets to obtain the set of targets at the current moment; associates and matches the set of targets at the current moment with the set of convex hull targets to update the contour state of the associated matching targets in the set of targets. Through the above target tracking method, the target bounding box result is combined with the convex hull technology set, and preliminary tracking is performed using the bounding box, and the true contour of the target is extracted through the convex hull, thereby improving the accuracy of contour description.
[0154] The target tracking method of the present application combines target bounding box detection and convex hull extraction to provide an innovative target contour tracking method. First, using the bounding box result output by the model and cooperating with Kalman filtering for target tracking can effectively capture the motion trajectory of the target in a dynamic environment. This method ensures that during the movement of the target, the system can update the position in real time, thereby maintaining continuous attention to the target.
[0155] The target tracking method of the present application further improves the accuracy of the contour by associating and matching the convex hull with the tracking target. When the convex hulls associated with the same target are clustered, a "measurement" of the true contour of the target at the current moment can be generated. Combining the description of the target true contour by the Gaussian process can effectively capture the shape change of the target and model it. This efficient contour update mechanism not only improves the tracking accuracy but also enhances the adaptability to environmental changes, enabling the system to accurately reflect the true contour of the target even in the case of occlusion or shape change. This series of technical measures makes the present invention show higher stability and reliability in the field of target contour tracking.
[0156] Those skilled in the art can understand that in the above method of the specific embodiment, the writing order of each step does not mean a strict execution order and does not constitute any limitation to the implementation process. The specific execution order of each step should be determined according to its function and possible internal logic.
[0157] To implement the above target tracking method, the present application also proposes a target tracking device. For details, please refer toFigure 5 , Figure 5 is a schematic structural diagram of an embodiment of the target tracking device provided by the present application.
[0158] The target tracking device 500 in this embodiment includes: an acquisition module 51, a detection module 52, a tracking module 53, and an update module 54.
[0159] Among them, the acquisition module 51 is used to acquire the laser point cloud at the current moment.
[0160] The detection module 52 is used to perform target detection on the laser point cloud to obtain a detection target set.
[0161] The detection module 52 is used to perform convex hull detection on the laser point cloud to obtain a convex hull target set.
[0162] The tracking module 53 is used to perform target tracking on the detection target set to obtain the target set at the current moment.
[0163] The update module 54 is used to perform association matching between the target set at the current moment and the convex hull target set to update the contour states of the associated and matched targets in the target set.
[0164] To implement the above target tracking method, the present application also proposes a target tracking device. For details, please refer to Figure 6 , Figure 6 is a schematic structural diagram of an embodiment of the target tracking device provided by the present application.
[0165] The target tracking device 400 in this embodiment includes a processor 41, a memory 42, an input / output device 43, and a bus 44.
[0166] The processor 41, the memory 42, and the input / output device 43 are respectively connected to the bus 44. Program data is stored in the memory 42, and the processor 41 is used to execute the program data to implement the target tracking method described in the above embodiment.
[0167] In an embodiment of the present application, the processor 41 may also be referred to as a CPU (Central Processing Unit). The processor 41 may be an integrated circuit chip with signal processing capabilities. The processor 41 may also be a general-purpose processor, a digital signal processor (DSP, Digital Signal Process), an application-specific integrated circuit (ASIC, Application Specific Integrated Circuit), a field-programmable gate array (FPGA, Field Programmable Gate Array), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor may be a microprocessor or the processor 41 may also be any conventional processor, etc.
[0168] The present application also provides a computer storage medium. Please continue to refer to Figure 7 , Figure 7 FIG. is a schematic structural diagram of an embodiment of the computer storage medium provided by the present application. A computer program 61 is stored in the computer storage medium 600. When the computer program 61 is executed by a processor, it is used to implement the target tracking method in the above embodiment.
[0169] When the embodiments of the present application are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions to enable a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.
[0170] The above are only the embodiments of the present application, and do not limit the patent scope of the present application. Any equivalent structural or equivalent process transformation made by using the content of the specification and drawings of the present application, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present application.
Claims
1. A target tracking method, characterized in that: The target tracking method comprises: Get the laser point cloud at the current moment; Performing target detection on the laser point cloud to obtain a detection target set; Performing convex hull detection on the laser point cloud to obtain a convex hull target set; Tracking the detected target set to obtain the target set at the current moment; The target set at the current moment is associated and matched with the convex hull target set to update the contour state of the associated and matched targets in the target set.
2. The target tracking method according to claim 1, characterized in that: The performing target tracking on the detection target set to obtain the target set at the current moment includes: Use the target set at the previous moment to predict the predicted target set at the current moment; Associatively matching the detected target set and the predicted target set at the current moment, and obtaining the detected target index of the associated matching target; The associated matching target is updated according to the detected target index to obtain the target set at the current moment.
3. The target tracking method according to claim 2, characterized in that: The associating and matching the detection target set and the predicted target set at the current moment to obtain the detection target index of the associated matching target includes: Constructing a cost matrix using the detection target boxes in the detection target set and the prediction target boxes in the prediction target set; Solving the cost matrix by cost minimization allocation; According to the solved value of the cost matrix, the detection target index of the associated matching target is determined.
4. The target tracking method according to claim 2 or 3, characterized in that: The updating of the associated matching target according to the detected target index to obtain the target set at the current moment includes: Obtaining a predicted target frame and a predicted covariance of the associated matching target; Obtaining an observed target frame of the predicted target frame using an observation matrix; Obtaining an innovation covariance using the observation matrix, the prediction covariance, and the measurement noise of the associated matching target; Obtaining a Kalman gain using the new information covariance, the observation matrix, and the prediction covariance; Using the Kalman gain, the detection target frame and the detection covariance of the associated matching target are updated; The target set at the current moment is obtained according to the updated detection target frame and detection covariance.
5. The target tracking method according to claim 1, characterized in that: The associating and matching the target set at the current moment with the convex hull target set includes: Based on each detection target frame of the target set at the current moment, traverse all convex hull targets in the convex hull target set to obtain an association value between the detection target frame and the convex hull target; The convex hull target index whose association value is greater than a preset association threshold is added to the association set of the association matching target.
6. The target tracking method according to claim 5, characterized in that: The updating of the contour status of the associated matching target in the target set includes: Merging and clustering the convex hull target point clouds in the associated set of the associated matching target to construct target contour measurement of the associated matching target; Based on the target contour state at the previous moment, obtain the predicted contour state at the current moment; Acquire a predicted measurement of the associated matching target using the position information of the detection target frame of the associated matching target and the predicted contour state; Based on the target covariance at the previous moment, obtain the predicted covariance at the current moment; Generate an innovation matrix using the predicted measurements and the target profile measurements; The predicted profile state is updated using the innovation matrix and the Kalman gain to obtain the profile state of the associated matching target; The predicted covariance is updated using the Kalman gain and the observation matrix to obtain the covariance of the associated matching target.
7. The target tracking method according to claim 6, characterized in that: The target tracking method further includes: Constructing an innovation covariance based on the prediction covariance and the equivalent measurement noise; The Kalman gain is constructed based on the prediction covariance, the innovation covariance and the observation matrix.
8. The target tracking method according to claim 6, characterized in that: The step of merging and clustering the convex hull target point clouds in the associated set of the associated matching target to construct the target contour measurement of the associated matching target includes: Merging and clustering the convex hull target point clouds in the associated set of the associated matching target to obtain a clustering result; According to the clustering result and the orientation angle of the detection target frame, a target contour measurement of the associated matching target is constructed.
9. A target tracking device, characterized in that: The target tracking device includes a memory and a processor coupled to the memory; The memory is used to store program data, and the processor is used to execute the program data to implement the target tracking method as described in any one of claims 1 to 8.
10. A computer storage medium, characterized in that: The computer storage medium is used to store program data, and when the program data is executed by a computer, it is used to implement the target tracking method according to any one of claims 1 to 8.