Three-dimensional multi-target tracking method and device based on life cycle hierarchical management

Abstract

The invention relates to the technical field of automatic driving environment perception, in particular to a three-dimensional multi-target tracking method and device based on life cycle hierarchical management. The method comprises the steps of preprocessing a three-dimensional target detection result, predicting a future state of a three-dimensional target tracking trajectory, matching and associating a detection frame and a trajectory frame at the same moment, and managing new and dead trajectories. Wherein the future state prediction of the three-dimensional target tracking trajectory adopts extended Kalman filtering based on interactive multi-model; and track newborn death management adopts a life cycle level-to-level management strategy. The method covers the maneuvering behaviors of constant-speed straight lines, acceleration and deceleration straight lines, constant-speed turning and acceleration and deceleration turning, follows a first principle that a close-distance target disappears more possibly to shield and a long-distance target disappears more possibly to leave a detection range, and more accurately and effectively predicts a future state and manages a life cycle. The method is beneficial to reducing identity switching, false positive and false negative tracks, and has excellent tracking performance.

Description

Technical Field

[0001] The present application relates to the field of autonomous driving environment perception technology, and more specifically, to a three-dimensional multi-target tracking method and device based on lifecycle hierarchical management. Background Art

[0002] Autonomous driving is an effective solution to relieve traffic congestion, improve driving safety, and enhance the driving experience. In an autonomous driving system, environmental perception provides information about the surrounding environment for decision-making planning and vehicle control. Multi-target tracking, as a key component of environmental perception, is used to obtain the speed attributes and identity of the target. The accuracy of multi-target tracking will directly affect the accuracy of trajectory prediction of the lower-level intelligent agent, and plays an important role in autonomous driving technology.

[0003] Continuous target space state information is indispensable for intelligent agent trajectory prediction. Compared with the joint detection multi-target tracking method, the detection-based multi-target tracking method splits the detection component and the tracking component to avoid the mutual constraints caused by excessive competition between the two tasks. However, the existing detection-based multi-target tracking method builds a Kalman filter based on a constant speed motion model, treats the target as uniform linear motion, and uses a fixed life cycle to manage the target trajectory, ignoring the influence of distance factors. There are a large number of identity switches, false positive and false negative trajectories, which are difficult to meet the performance requirements of autonomous driving.

[0004] Therefore, it is very necessary to design a three-dimensional multi-target tracking method with life cycle hierarchical management covering uniform straight line, accelerated and decelerated straight line, uniform speed turning, and accelerated and decelerated turning motions. Summary of the invention

[0005] In order to solve the problem that the existing detection-based multi-target tracking method constructs a Kalman filter based on a constant speed motion model, regards the target as uniform linear motion, adopts a fixed life cycle management target trajectory, ignores the influence of distance factors, has a large number of identity switches, false positive and false negative trajectories, and is difficult to meet the requirements of autonomous driving performance, the present application provides a three-dimensional multi-target tracking method and device based on life cycle hierarchical management.

[0006] The embodiment of the present application is implemented as follows:

[0007] In a first aspect, the present application provides a three-dimensional multi-target tracking method based on lifecycle hierarchical management, comprising:

[0008] Input the three-dimensional target detection result at time T;

[0009] The three-dimensional target detection results at time T are sequentially subjected to score filtering and non-maximum suppression to filter out low-quality detection frames and obtain the preprocessed detection frames at time T;

[0010] Read the activation status track at time T-1;

[0011] According to the activation state trajectory at time T-1, the weighted trajectory frame at time T is obtained through the Kalman filter;

[0012] The cost matrix between the preprocessed T-time detection frame and the weighted T-time trajectory frame is calculated using the 3D distance intersection and union ratio;

[0013] According to the cost matrix between the preprocessed T-time detection frame and the weighted T-time trajectory frame, the preprocessed T-time detection frame and the weighted T-time trajectory frame are associated using the bipartite graph-based Hungarian matching to obtain unmatched trajectories, unmatched detections, and matched trajectory detection pairs;

[0014] The detection range is divided into N parts in a circular shape, and the circular distance areas from small to large correspond to the maximum number of matching failures K1, K2,,,,,,,,KN;

[0015] Calculate the bird's-eye view distance of the center point of the trajectory frame based on the unmatched trajectory, and determine the distance range to which the center point of the trajectory frame belongs;

[0016] Initialize a fixed minimum number of successful matches and a maximum number of failed matches based on the distance range;

[0017] Obtaining an activated state trajectory based on the unmatched trajectory, the unmatched detection and the matched trajectory detection;

[0018] Access the fused trajectory box in the active state trajectory and output the three-dimensional target speed and identity at time T.

[0019] In a possible implementation, the three-dimensional target detection result is (x d ,y d ,z d ,l d ,w d ,h d ,θ d ,s d ), where x d ,y d 、z d Indicates the center coordinate of the detection box, l d 、w d 、h d Represents the length, width and height of the detection box, θ d Indicates the angle of the detection box, s d Indicates the confidence of the detection box.

[0020] In a possible implementation, the activation state trajectory is where x t ,y t 、zt Indicates the coordinates of the center point of the trajectory box, l t 、w t 、h t Indicates the length, width and height of the trajectory box, v x_t 、v y_t v z_t Represents the velocity of the trajectory frame, v t represents the combined velocity of the trajectory frame, a x_t 、a y_t 、a z_t represents the acceleration of the trajectory frame, a t represents the total acceleration of the trajectory frame, θ t Indicates the angle of the trajectory box, Indicates the angle change rate of the trajectory frame, s t It represents the confidence of the track box, and ID represents the identity of the track box.

[0021] In a possible implementation, the step of obtaining a weighted trajectory frame at time T by using a Kalman filter according to the activation state trajectory at time T-1 further includes:

[0022] The four extended Kalman filter activation state trajectories at time T-1 are multiplied by the corresponding extended Kalman filter interaction probabilities to obtain four interactive T-1 time activation state trajectories. The four interactive T-1 time activation state trajectories are respectively predicted to obtain four estimated T time activation state trajectories. The four estimated T time activation state trajectories are multiplied by the corresponding extended Kalman filter probabilities and then added together to obtain the weighted T time trajectory frame.

[0023] In a possible implementation, obtaining the trajectory of the activation state based on the unmatched trajectory, the unmatched detection and the matched trajectory detection further includes:

[0024] For an unmatched track, if the number of consecutive matching failures is greater than the maximum number of matching failures based on the distance range, the track is deleted; otherwise, the track is not deleted and is set to the active state.

[0025] In a possible implementation, obtaining the trajectory of the activation state based on the unmatched trajectory, the unmatched detection and the matched trajectory detection further includes:

[0026] For unmatched detection, if the number of consecutive successful matches is less than the fixed minimum number of successful matches, no track is created; otherwise, a track is created and set to the active state.

[0027] In a possible implementation, obtaining the trajectory of the activation state based on the unmatched trajectory, the unmatched detection and the matched trajectory detection further includes:

[0028] For the matching trajectory detection pair, the measurement state is used to update the state of the four estimated activation state trajectories at time T respectively to obtain four revised activation state trajectories at time T, and the extended Kalman filter interaction probability and the extended Kalman filter probability are updated. The four revised activation state trajectories at time T are multiplied by the corresponding extended Kalman filter probabilities and then added to obtain the fused trajectory frame at time T, and the trajectory is set to the activation state.

[0029] In one possible implementation, the calculation formula of the three-dimensional distance intersection ratio is: Where DIoU represents the 3D distance intersection over union, IoU represents the 3D intersection over union, d represents the Euclidean distance between the center points of the detection box and the trajectory box, and c represents the diagonal length of the smallest cuboid containing the detection box and the trajectory box.

[0030] In one possible implementation, the calculation formula for the bird's-eye view distance of the center point of the trajectory box is: Where L BEV Indicates the bird's-eye view distance of the center point of the trajectory box, L BEV_x Indicates the distance of the center point of the trajectory box along the X axis, L BEV_y Indicates the distance of the center point of the track box along the Y axis.

[0031] In a second aspect, the present application provides a three-dimensional multi-target tracking device based on lifecycle hierarchical management, comprising:

[0032] A data input module, used to input three-dimensional target detection results;

[0033] A preprocessing module, used for performing score filtering and non-maximum suppression on the three-dimensional target detection result to obtain a detection frame at time T;

[0034] The state prediction module is used to obtain the weighted T-time prediction frame through extended Kalman filtering according to the activation state trajectory;

[0035] The data association module is used to obtain unmatched trajectories, unmatched detections, and matched trajectory detection pairs based on the detection box at time T through Hungarian matching with the three-dimensional distance intersection ratio as the similarity measurement;

[0036] A trajectory management module, used to manage the unmatched trajectory, unmatched detection and matched trajectory detection pairs to obtain an activated state trajectory;

[0037] The data output module is used to output the speed and identity of the three-dimensional target.

[0038] The technical solution provided by this application can at least achieve the following beneficial effects:

[0039] The three-dimensional multi-target tracking method and device based on lifecycle hierarchical management provided in the present application include two types of hyperparameters, namely, a fixed minimum number of successful matches and a maximum number of failed matches based on a distance range, through a lifecycle hierarchical management strategy. At the same time, the motion model and lifecycle management strategy are redesigned according to the maneuvering behavior and the distance range. The improved algorithm predicts the future state more accurately and manages the lifecycle more effectively. The algorithm is conducive to reducing identity switching, false positive and false negative trajectories, and has excellent tracking performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, a brief introduction will be given below to the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0041] Figure 1 It is a flowchart of a three-dimensional multi-target tracking method based on life cycle hierarchical management shown in an exemplary embodiment of the present application;

[0042] Figure 2 It is a schematic diagram of a specific implementation process of a tracking method shown in an exemplary embodiment of the present application;

[0043] Figure 3 is a schematic diagram of an extended Kalman filter based on an interactive multi-model according to an exemplary embodiment of the present application;

[0044] Figure 4 is a schematic diagram of a three-dimensional distance intersection-to-union ratio shown in an exemplary embodiment of the present application;

[0045] Figure 5 It is a schematic diagram of the corresponding relationship between the distance range and the maximum number of matching failures in the life cycle hierarchical management strategy shown in an exemplary embodiment of the present application;

[0046] Figure 6 It is a structural schematic diagram of a three-dimensional multi-target tracking device based on lifecycle hierarchical management shown in an exemplary embodiment of the present application.

[0047] Reference numerals:

[0048] 1. Data input module; 2. Preprocessing module; 3. State prediction module; 4. Data association module; 5. Trajectory management module; 6. Data output module. DETAILED DESCRIPTION

[0049] In order to make the purpose, implementation mode and advantages of the present application clearer, the exemplary implementation mode of the present application will be clearly and completely described below in conjunction with the drawings in the exemplary embodiments of the present application. Obviously, the described exemplary embodiments are only part of the embodiments of the present application, not all of the embodiments. It should be understood that the specific embodiments described here are only used to explain the present application and are not used to limit the present application.

[0050] It should be noted that the brief description of terms in this application is only for the convenience of understanding the embodiments described below, and is not intended to limit the embodiments of this application. Unless otherwise specified, these terms should be understood according to their common and usual meanings.

[0051] The terms "first", "second", "third", etc. in the specification and claims of this application and the above drawings are used to distinguish similar or similar objects or entities, and do not necessarily mean to limit a specific order or sequence, unless otherwise noted. It should be understood that the terms used in this way can be interchangeable under appropriate circumstances.

[0052] The terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover but not exclude inclusion, for example, a product or device comprising a list of components is not necessarily limited to all the components expressly listed but may include other components not expressly listed or inherent to such product or device.

[0053] Before explaining the three-dimensional multi-target tracking method based on lifecycle hierarchical management provided in the embodiment of the present application, the application scenario and implementation environment of the embodiment of the present application are first introduced.

[0054] Autonomous driving is an effective solution to relieve traffic congestion, improve driving safety, and enhance the driving experience. In an autonomous driving system, environmental perception provides information about the surrounding environment for decision-making planning and vehicle control. Multi-target tracking, as a key component of environmental perception, is used to obtain the speed attributes and identity of the target. The accuracy of multi-target tracking will directly affect the accuracy of trajectory prediction of the lower-level intelligent agent, and plays an important role in autonomous driving technology.

[0055] Continuous target space state information is indispensable for intelligent agent trajectory prediction. Compared with the joint detection multi-target tracking method, the detection-based multi-target tracking method splits the detection component and the tracking component to avoid the mutual constraints caused by excessive competition between the two tasks. However, the existing detection-based multi-target tracking method builds a Kalman filter based on a constant speed motion model, treats the target as uniform linear motion, and uses a fixed life cycle to manage the target trajectory, ignoring the influence of distance factors. There are a large number of identity switches, false positive and false negative trajectories, which are difficult to meet the performance requirements of autonomous driving.

[0056] Therefore, it is very necessary to design a three-dimensional multi-target tracking method with life cycle hierarchical management covering uniform straight line, accelerated and decelerated straight line, uniform speed turning, and accelerated and decelerated turning motions.

[0057] Based on this, this application provides a three-dimensional multi-target tracking method and device based on life cycle hierarchical management, redesigning the motion model and life cycle management strategy according to the maneuvering behavior and distance range. The improved algorithm predicts the future state more accurately and manages the life cycle more effectively. The algorithm is conducive to reducing identity switching, false positive and false negative trajectories, and has excellent tracking performance.

[0058] Next, the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems will be described in detail through embodiments and in combination with the accompanying drawings. The embodiments may be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. Obviously, the described embodiments are part of the embodiments of the present application, not all of them.

[0059] Figure 1 It is a flowchart of a three-dimensional multi-target tracking method based on lifecycle hierarchical management shown in an exemplary embodiment of the present application.

[0060] In an exemplary embodiment, Figure 1 As shown, a three-dimensional multi-target tracking method based on life cycle hierarchical management is provided, and the method may include the following steps:

[0061] Step 01: Input the 3D target detection result at time T;

[0062] Step 02: Perform score filtering and non-maximum suppression on the three-dimensional target detection results at time T in turn, filter out low-quality detection frames, and obtain the preprocessed detection frame at time T;

[0063] Step 03: Read the activation status track at time T-1;

[0064] Step 04: Based on the activation state trajectory at time T-1, the weighted trajectory frame at time T is obtained through the Kalman filter;

[0065] Step 05: Use the 3D distance intersection ratio to calculate the cost matrix between the preprocessed T-time detection frame and the weighted T-time trajectory frame;

[0066] Step 06: According to the cost matrix between the preprocessed T-time detection frame and the weighted T-time trajectory frame, the preprocessed T-time detection frame and the weighted T-time trajectory frame are associated using the bipartite graph-based Hungarian matching method to obtain unmatched trajectories, unmatched detections, and matched trajectory detection pairs;

[0067] Step 07: Divide the detection range into N parts in a circular shape, and the circular distance areas correspond to the maximum number of matching failures K1, K2,,,,,,,,,KN from small to large;

[0068] Step 08: Calculate the bird's-eye view distance of the center point of the trajectory frame based on the unmatched trajectory, and determine the distance range to which the center point of the trajectory frame belongs;

[0069] Step 09: Initialize a fixed minimum number of successful matches and a maximum number of failed matches based on the distance range;

[0070] Step 10: obtaining an activated trajectory based on the unmatched trajectory, the unmatched detection and the matched trajectory detection;

[0071] Step 11: Access the fused trajectory frame in the active state trajectory and output the three-dimensional target speed and identity at time T.

[0072] Figure 2 is a schematic diagram of a specific implementation flow of a tracking method shown in an exemplary embodiment of the present application, Figure 3 is a schematic diagram of an extended Kalman filter based on an interactive multi-model according to an exemplary embodiment of the present application, Figure 4 is a schematic diagram of a three-dimensional distance intersection-to-union ratio shown in an exemplary embodiment of the present application, Figure 5 It is a schematic diagram of the correspondence between the distance range and the maximum number of matching failures in the lifecycle hierarchical management strategy shown in an exemplary embodiment of the present application.

[0073] In one possible implementation, Figure 2 As shown, the three-dimensional multi-target tracking based on lifecycle hierarchical management proposed in this application is specifically implemented by the following steps:

[0074] S1, input the 3D target detection result (x d ,y d ,z d ,l d ,w d ,h d ,θ d ,s d ), where x d ,y d 、z d Indicates the center coordinate of the detection box, l d 、w d 、h d Represents the length, width and height of the detection box, θ d Indicates the angle of the detection box, s d Indicates the confidence of the detection box.

[0075] S2, the three-dimensional target detection results at time T are sequentially subjected to score filtering and non-maximum suppression to filter out low-quality detection frames and obtain the preprocessed detection frames at time T.

[0076] S3, read the activation status track at time T-1 where x t ,y t 、z t Indicates the coordinates of the center point of the trajectory box, l t 、w t 、h t Indicates the length, width and height of the trajectory box, v x_t 、v y_t v z_t Represents the velocity of the trajectory frame, v t represents the combined velocity of the trajectory frame, a x_t 、a y_t 、a z_t represents the acceleration of the trajectory frame, a t represents the total acceleration of the trajectory frame, θ t Indicates the angle of the trajectory box, Indicates the angle change rate of the trajectory frame, s t It represents the confidence of the track box, and ID represents the identity of the track box.

[0077] S4, multiply the four extended Kalman filter activation state trajectories at time T-1 by the corresponding extended Kalman filter interaction probabilities to obtain four interactive T-1 activation state trajectories, perform state prediction on the four interactive T-1 activation state trajectories respectively to obtain four estimated T time activation state trajectories, multiply the four estimated T time activation state trajectories by the corresponding extended Kalman filter probabilities and then add them to obtain a weighted T time trajectory frame. The extended Kalman filter based on interactive multi-models is composed of an extended Kalman filter based on a constant speed motion model, an extended Kalman filter based on a constant acceleration motion model, an extended Kalman filter based on a constant turning rate and speed motion model, and an extended Kalman filter based on a constant turning rate and acceleration motion model. The extended Kalman filter based on interactive multi-models is as follows: Figure 3 shown.

[0078] S5, using the 3D distance intersection and union ratio to calculate the cost matrix between the preprocessed T-time detection frame and the weighted T-time trajectory frame. The calculation formula of the 3D distance intersection and union ratio is: Where DIoU represents the 3D distance intersection over union, IoU represents the 3D intersection over union, d represents the Euclidean distance between the center points of the detection box and the trajectory box, c represents the diagonal length of the smallest cuboid containing the detection box and the trajectory box, and the 3D distance intersection over union is Figure 4 shown.

[0079] S6, according to the cost matrix between the preprocessed T-time detection frame and the weighted T-time trajectory frame, the preprocessed T-time detection frame and the weighted T-time trajectory frame are associated using the bipartite graph-based Hungarian matching to obtain unmatched trajectories, unmatched detections, and matched trajectory detection pairs.

[0080] S7, the detection range is divided into N parts in a ring shape. The ring distance areas correspond to the maximum matching failure times K1, K2, ······, KN from small to large. In order to meet the first principle that close-range targets are more likely to be blocked when they disappear, and long-range targets are more likely to leave the detection range when they disappear, the maximum matching failure times satisfy K1>K2>······>KN. The corresponding relationship between the distance range and the maximum matching failure times in the life cycle hierarchical management strategy is as follows: Figure 5 shown.

[0081] S8, calculating the bird's-eye view distance of the center point of the trajectory frame based on the unmatched trajectory, and determining the distance range to which the center point of the trajectory frame belongs. The bird's-eye view distance calculation formula of the center point of the trajectory frame is: Where L BEV Indicates the bird's-eye view distance of the center point of the trajectory box, L BEV_x Indicates the distance of the center point of the trajectory box along the X-axis direction, L BEV_y Indicates the distance of the center point of the track box along the Y axis.

[0082] S9, initializing a fixed minimum number of successful matches and a maximum number of failed matches based on a distance range.

[0083] S10, for an unmatched trajectory, if the number of consecutive matching failures is greater than the maximum number of matching failures based on the distance range, then the trajectory is deleted; otherwise, the trajectory is not deleted and is set to an activated state.

[0084] S11, for unmatched detection, if the number of consecutive successful matches is less than a fixed minimum number of successful matches, then no track is created; otherwise, a track is created and set to an active state.

[0085] S12, for the matching trajectory detection pair, use the measurement state to update the state of the four estimated T-time activation state trajectories respectively to obtain four revised T-time activation state trajectories, update the extended Kalman filter interaction probability and the extended Kalman filter probability, multiply the four revised T-time activation state trajectories by the corresponding extended Kalman filter probabilities and then add them to obtain the fused T-time trajectory frame, and set the trajectory to the activation state.

[0086] S13, access the fused trajectory frame in the activated state trajectory, and output the three-dimensional target speed and identity at time T.

[0087] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the instructions, these steps are not necessarily executed in sequence according to the order of the instructions. Unless there is a clear explanation in this article, the execution of these steps is not strictly limited in order, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the above-mentioned embodiments may include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of the steps or stages in other steps.

[0088] Corresponding to the aforementioned embodiment of the three-dimensional multi-target tracking method based on lifecycle hierarchical management, adopting the same technical concept, the present application also provides an embodiment of a three-dimensional multi-target tracking device based on lifecycle hierarchical management.

[0089] Figure 6 It is a structural schematic diagram of a three-dimensional multi-target tracking device based on lifecycle hierarchical management shown in an exemplary embodiment of the present application.

[0090] In an exemplary embodiment, Figure 6 As shown, the three-dimensional multi-target tracking device based on life cycle hierarchical management includes:

[0091] Data input module 1, used to input three-dimensional target detection results;

[0092] Preprocessing module 2, used for performing score filtering and non-maximum suppression on the three-dimensional target detection result to obtain a detection frame at time T;

[0093] The state prediction module 3 is used to obtain a weighted T-time prediction frame through an extended Kalman filter according to the activation state trajectory;

[0094] The data association module 4 is used to obtain the unmatched trajectory 4, the unmatched detection and the matched trajectory detection pair according to the detection frame at time T through Hungarian matching with the three-dimensional distance intersection ratio as the similarity measurement;

[0095] A track management module 5 is used to manage the unmatched track, the unmatched detection and the matched track detection pairs to obtain an activated state track;

[0096] The data output module 6 is used to output the speed and identity of the three-dimensional target.

[0097] For the specific limitations of the three-dimensional multi-target tracking device based on lifecycle hierarchical management, please refer to the limitations of the three-dimensional multi-target tracking method based on lifecycle hierarchical management above, which will not be repeated here. Each module in the above-mentioned three-dimensional multi-target tracking device based on lifecycle hierarchical management can be implemented in whole or in part through 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.

[0098] 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.

[0099] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A three-dimensional multi-target tracking method based on life cycle hierarchical management, characterized in that: include: Input the three-dimensional target detection result at time T; The three-dimensional target detection results at time T are sequentially subjected to score filtering and non-maximum suppression to filter out low-quality detection frames and obtain the preprocessed detection frames at time T; Read the activation status track at time T-1; According to the activation state trajectory at time T-1, the weighted trajectory frame at time T is obtained through the Kalman filter; The cost matrix between the preprocessed T-time detection frame and the weighted T-time trajectory frame is calculated using the 3D distance intersection and union ratio; According to the cost matrix between the preprocessed T-time detection frame and the weighted T-time trajectory frame, the preprocessed T-time detection frame and the weighted T-time trajectory frame are associated using the bipartite graph-based Hungarian matching to obtain unmatched trajectories, unmatched detections, and matched trajectory detection pairs; The detection range is divided into N parts in a circular shape, and the circular distance areas from small to large correspond to the maximum number of matching failures K1, K2,,,,,,,,KN; Calculate the bird's-eye view distance of the center point of the trajectory frame based on the unmatched trajectory, and determine the distance range to which the center point of the trajectory frame belongs; Initialize a fixed minimum number of successful matches and a maximum number of failed matches based on the distance range; Obtaining an activated state trajectory based on the unmatched trajectory, the unmatched detection and the matched trajectory detection; Access the fused trajectory box in the active state trajectory and output the three-dimensional target speed and identity at time T.

2. The three-dimensional multi-target tracking method based on lifecycle hierarchical management according to claim 1, characterized in that: The three-dimensional target detection result is.(x d ,y d ,z d ,l d ,w d ,h d ,θ d ,s d ), where x d ,y d 、z d Indicates the center coordinate of the detection box, l d 、.w d .、h d Represents the length, width and height of the detection box, θ d Indicates the angle of the detection box, s d Indicates the confidence of the detection box.

3. The three-dimensional multi-target tracking method based on lifecycle hierarchical management according to claim 1, characterized in that: The activation state trajectory is where x t ,y t 、z t Indicates the coordinates of the center point of the trajectory box. t .、w t 、h t Indicates the length, width and height of the trajectory box, v x_t 、v y_t v z_t Represents the velocity of the trajectory frame, v t represents the combined velocity of the trajectory frame, a x_t .a y_t .、a z_t represents the acceleration of the trajectory frame, a t represents the total acceleration of the trajectory frame, θ t Indicates the angle of the trajectory box, Indicates the angle change rate of the trajectory frame, s t It represents the confidence of the track box, and ID represents the identity of the track box.

4. The three-dimensional multi-target tracking method based on lifecycle hierarchical management according to claim 1, characterized in that: The weighted T-time trajectory frame is obtained by using a Kalman filter according to the activation state trajectory at time T-1, further comprising: The four extended Kalman filter activation state trajectories at time T-1 are multiplied by the corresponding extended Kalman filter interaction probabilities to obtain four interactive T-1 time activation state trajectories. The four interactive T-1 time activation state trajectories are respectively predicted to obtain four estimated T time activation state trajectories. The four estimated T time activation state trajectories are multiplied by the corresponding extended Kalman filter probabilities and then added together to obtain the weighted T time trajectory frame.

5. The three-dimensional multi-target tracking method based on lifecycle hierarchical management according to claim 1, characterized in that: The step of obtaining the activated state trajectory based on the unmatched trajectory, the unmatched detection and the matched trajectory detection further comprises: For an unmatched track, if the number of consecutive matching failures is greater than the maximum number of matching failures based on the distance range, the track is deleted; otherwise, the track is not deleted and is set to the active state.

6. The three-dimensional multi-target tracking method based on lifecycle hierarchical management according to claim 5, characterized in that: The step of obtaining the activated state trajectory based on the unmatched trajectory, the unmatched detection and the matched trajectory detection further comprises: For unmatched detection, if the number of consecutive successful matches is less than the fixed minimum number of successful matches, no track is created; otherwise, a track is created and set to the active state.

7. The three-dimensional multi-target tracking method based on lifecycle hierarchical management according to claim 6, characterized in that: The step of obtaining the activated state trajectory based on the unmatched trajectory, the unmatched detection and the matched trajectory detection further comprises: For the matching trajectory detection pair, the measurement state is used to update the state of the four estimated activation state trajectories at time T respectively to obtain four revised activation state trajectories at time T, and the extended Kalman filter interaction probability and the extended Kalman filter probability are updated. The four revised activation state trajectories at time T are multiplied by the corresponding extended Kalman filter probabilities and then added to obtain the fused trajectory frame at time T, and the trajectory is set to the activation state.

8. The three-dimensional multi-target tracking method based on lifecycle hierarchical management according to claim 1, characterized in that: The calculation formula of the three-dimensional distance intersection ratio is: Where DIoU represents the 3D distance intersection over union, IoU represents the 3D intersection over union, d represents the Euclidean distance between the center points of the detection box and the trajectory box, and c represents the diagonal length of the smallest cuboid containing the detection box and the trajectory box.

9. The three-dimensional multi-target tracking method based on lifecycle hierarchical management according to claim 1, characterized in that: The calculation formula of the bird's-eye view distance of the center point of the trajectory box is: Where L BEV Indicates the bird's-eye view distance of the center point of the trajectory box, L BEVx Indicates the distance of the center point of the trajectory box along the X axis, L BEVy Indicates the distance of the center point of the track box along the Y axis.

10. A three-dimensional multi-target tracking device based on life cycle hierarchical management, characterized in that: include: A data input module, used to input three-dimensional target detection results; A preprocessing module, used for performing score filtering and non-maximum suppression on the three-dimensional target detection result to obtain a detection frame at time T; The state prediction module is used to obtain the weighted T-time prediction frame through extended Kalman filtering according to the activation state trajectory; The data association module is used to obtain unmatched trajectories, unmatched detections, and matched trajectory detection pairs based on the detection box at time T through Hungarian matching with the three-dimensional distance intersection ratio as the similarity measurement; A trajectory management module, used to manage the unmatched trajectory, unmatched detection and matched trajectory detection pairs to obtain an activated state trajectory; The data output module is used to output the speed and identity of the three-dimensional target.

Images