Multi-target tracking method, storage medium and electronic equipment
The target features are extracted and updated through deep neural networks and particle filtering methods, and the problem of degradation of target tracking accuracy in complex scenarios is solved, achieving higher target matching accuracy and tracking stability.
Patent Information
- Application Number
- CN202411926087.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-25
- Publication Date
- 2025-04-18
AI Technical Summary
In complex scenarios, target features are prone to missing, resulting in a decrease in tracking accuracy. Especially in low-resolution video, bad weather and complex backgrounds, existing multi-target tracking technologies are difficult to effectively solve these problems.
Deep neural network is used to extract target features, create and update the target motion model through particle filtering method, combine the deep appearance feature extraction network to calculate the similarity between particles and input observations, and improve the accuracy of target matching.
It significantly improves the matching accuracy of the detection results and existing targets during the target tracking and trajectory generation process in complex scenarios, and improves the performance and stability of multi-objective tracking technology.
Smart Images
Figure CN119850676A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of computer vision technology, and in particular to a multi-target tracking method, a storage medium and an electronic device. Background Art
[0002] In today's era, target tracking has become one of the most popular research hotspots in the field of computer vision. With the vigorous development of artificial intelligence technology, the continuous improvement of hardware equipment performance and the rapid rise of deep network technology, the performance of target tracking in practical applications has been greatly enhanced, and related research can tailor more optimized algorithms and model structures according to specific scenarios. At present, multi-target tracking technology has gradually penetrated into many fields such as traffic safety, virtual reality and unmanned driving.
[0003] In the field of traffic safety, given the increasing number of cars and the growing pressure on road transportation, the application of multi-target tracking technology provides an effective means for monitoring vehicle violations. Multi-target tracking technology can accurately capture the driving status of each vehicle and detect violations in a timely manner; in the field of virtual reality research, virtual scenes often cover a variety of people or vehicle entities. With the help of target tracking technology, each entity can be given a certain degree of autonomous decision-making ability, making its behavior more intelligent and realistic; in the field of unmanned driving, although there are still many challenges in achieving L4 or even L5 levels of highly automated driving, the application examples of L2 level unmanned vehicles and fixed-route unmanned trucks launched by Tesla have already allowed people to see the dawn of unmanned driving technology becoming a reality. In this process, multi-target tracking technology plays a key role in helping vehicles identify surrounding traffic participants and ensure driving safety.
[0004] However, there are still many thorny problems in the field of multi-target tracking research. For example, in scenes with complex backgrounds, it is common for various objects in the field of view to block each other, resulting in the loss of target features, which greatly interferes with tracking accuracy; for example, in the face of low-resolution videos, people or objects in the long-distance perspective are easily confused with the background, adding many difficulties to the target detection task; in addition, such as different lighting conditions caused by the alternation of day and night, and bad weather scenes such as rain and snow, time and weather factors constantly test the performance of multi-target tracking algorithms, becoming a bottleneck restricting the further development of multi-target tracking technology. Summary of the invention
[0005] The purpose of the present invention is to provide a multi-target tracking method, a storage medium and an electronic device, which have the advantage of strong multi-target tracking capability.
[0006] To achieve the above object, the present invention provides a multi-target tracking method, which comprises:
[0007] S10, obtaining picture information of the current frame, extracting target information to be matched of the current frame, matching the target information to be matched with existing targets extracted from previous frames, and obtaining successfully matched targets and unmatched targets;
[0008] S20, for the unmatched target, determining whether it is a new target or a disappeared target;
[0009] S30, for a new target, creating a particle filter for describing the new target; for a disappeared target, deleting an existing particle filter for describing the disappeared target;
[0010] S40: Output trajectory tracking results.
[0011] Optionally, step S10 includes:
[0012] S101. Create motion models for each particle Where [x, y] is the coordinate of the center point of the bbox bounding box. is the movement speed of the bbox bounding box, [h, w] is bbo x The length and width of the bounding box;
[0013] S102, maintaining the motion state of each particle, and updating the particle motion state when the next a posteriori knowledge arrives. The kinematic model of the particle is expressed as: In the formula, x k represents the state of the system at time k, y k represents the observed value of the system at time k, f k and h k They represent the state transfer function and system measurement function of the system, u k and v k They represent the system noise and observation noise of the system at time k respectively;
[0014] S103, based on the observed values and motion states from time 1 to time k-1, predict the motion state of the system at time k, that is, from p(x k-1 |y 1：k-1 ) to obtain p(x k |y k-1 ), given x k-1 , according to the system-related assumptions, the state xk and y k-1 Independent of each other, we can get: x for the upper formula k-1 Integrate and finally get the corresponding credibility expression: p(x k |y k-1 )=∫p(x k |x k-1 )p(x k-1 |yk-1 )dx k-1 ;
[0015] S104, update the observed value of the system at time k, that is, update p(xk|y k-1 ) to get p(x k |y k ) process, the predicted value x at time k is known k , we can get the posterior probability: Where p(y k |y k-1 ) can be taken as a normalized constant in actual calculations. The specific calculation process is: p(y k |y k-1 )=∫p(y k |x k )p(x k |Y k-1 )dx k ;
[0016] S105, constructing a deep appearance feature extraction network for particle filtering through a deep neural network structure of a depthwise over-parameterized convolutional layer (DO-Conv), and using the appearance features extracted by the deep appearance feature extraction network to calculate the weights of the particles and affect the subsequent importance ranking;
[0017] S106. Calculate the similarity d between particle i and input observation i .
[0018] Optionally, the input image size of the deep appearance feature extraction network is 128×64, and a 128-dimensional feature vector is finally obtained by setting the network layer structure.
[0019] Optionally, in S106, according to the formula Get the similarity d between particle i and the input observation i , where is the deep appearance feature vector of the input observation, r i It is the feature vector corresponding to a particle in the particle filter system at the current moment.
[0020] Optionally, step S20 specifically includes:
[0021] S201, according to the similarity d i Determine whether the observed value of the particle is reasonable;
[0022] S202. If the observed value of the particle is unreasonable, the predicted value of the particle is used to transfer the motion state.
[0023] Optionally, step S20 further includes S203, setting a target state maintenance model, adding auxiliary variables to each target to record the existence of the target, and the target state maintenance model is: X t =[(x 1，t , e 1，t ),…,(x j，t ), e j,t ], j∈(1, 2, 3, …, M t ), where M t is the number of states of the tracking system at time t; x j，t represents the state of the jth target at time t; e j，t Represents x j，t The state of existence at time t; when e j，t =1, indicating x j，t Existence; when e j，t =0, indicating x j，t does not exist; the number of targets at time t can be obtained by Calculated.
[0024] Optionally, step S203 further includes: using Y 1：t ={Y1, ..., Y t} represents all observations of the tracking system at time t and before, where Y k ={x i,k |k=1,…,t,t=1,…,N k}, where Y k is the i-th observation value at the k-th time, N k is the number of observations at the kth moment, given the known observation value Y 1：t When , the state of each target can be inferred based on the particle filter.
[0025] Optionally, in step S201, the judgment index b is calculated, and the judgment index b is calculated according to the following formula: When b is greater than the threshold t2, the current prediction value is considered reasonable.
[0026] Optionally, in step S202, using the predicted value of the particle to transfer the motion state includes: Where x k represents the motion state of the particle, w k Represents the normalized weight of the particle.
[0027] Optionally, in step S30, for a new target, creating a particle filter for describing the new target includes: determining whether the unmatched target is continuously in T newIf the match is successful in the frame video, it is determined to be a new target, and a particle filter for describing the new target is created.
[0028] Optionally, in step S30, for a disappeared target, deleting the existing particle filter describing the disappeared target includes: determining whether the unmatched target is continuously in T dead If the matching fails in the frame video, the target is determined to have disappeared, and the particle filter used to describe the disappeared target is deleted.
[0029] Optionally, after step S10, the method further includes step S11, updating the particle filter state of the successfully matched target for the successfully matched target, and proceeding to S40.
[0030] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the multi-target tracking method as described above is implemented.
[0031] The present invention also provides an electronic device, including a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the multi-target V-tracking method as described above is implemented.
[0032] In summary, compared with the prior art, the multi-target tracking method, storage medium and electronic device provided by the present invention have the following beneficial effects:
[0033] The multi-target tracking method of the present invention uses a deep neural network to extract target features, performs similarity calculation on the corresponding target depth features, and uses the idea of multi-level matching to improve the accuracy of target matching, thereby obtaining a more effective registration model. The multi-target tracking method of the present invention can significantly improve the matching accuracy of detection results and existing targets in the process of target tracking and trajectory generation in complex scenarios, and plays an important role in the fields of passenger flow analysis, behavior modeling, and unmanned driving. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 It is an overall flow chart of the multi-target tracking method of the present invention.
[0035] Figure 2 Flowchart of the particle filtering method in the multi-target tracking method.
[0036] Figure 3 Schematic diagram of the residual block of the deep neural network in the present invention.
[0037] Figure 4 Schematic diagram of the feature extraction network of the present invention.
[0038] Figure 5Schematic diagram of the process of maintaining target status in the multi-target tracking method.
[0039] Figure 6 Schematic diagram of the new target judgment process in the multi-target tracking method.
[0040] Figure 7 Schematic diagram of the disappearing target judgment process in the multi-target tracking method. DETAILED DESCRIPTION
[0041] The following will be combined with the attached embodiment of the present invention Figure 1 ~Attached Figure 7 , the technical solutions, structural features, objectives achieved and effects in the embodiments of the present invention are described in detail.
[0042] It should be noted that the drawings are in a very simplified form and use non-precise proportions. They are only used to conveniently and clearly assist in explaining the embodiments of the present invention, and are not used to limit the conditions for the implementation of the present invention. Therefore, they have no substantive technical significance. Any structural modification, change in proportional relationship or adjustment of size, without affecting the effects and purposes that can be achieved by the present invention, should still fall within the scope of the technical content disclosed by the present invention.
[0043] It should be noted that, in the present invention, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only the elements explicitly listed, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.
[0044] like Figure 1 As shown, the present invention provides a multi-target tracking method, comprising:
[0045] S10, obtaining the image information of the current frame, extracting the target information to be matched of the current frame, matching the target information to be matched with the existing targets extracted from the previous frames, and obtaining the successfully matched targets and the unmatched successfully matched targets. In this step, the target information to be matched is extracted from the image information of the current frame, and the target information to be matched is matched with the existing targets extracted from the previous frames, and obtaining the successfully matched targets and the unmatched successfully matched targets, wherein the successfully matched targets are classified into the matched trajectory set, and the unmatched successfully matched targets are classified into the unmatched detection target set.
[0046] S20: For the unmatched target, use the trajectory maintenance module to determine whether it is a new target or a disappeared target.
[0047] S30, for a new target, creating a particle filter for describing the new target; for a disappeared target, deleting the existing particle filter for describing the disappeared target;
[0048] S40: Output trajectory tracking results.
[0049] like Figure 2 As shown, in this embodiment, step S10 specifically includes:
[0050] S101. Establish a unified motion model for the samples in the particle filter method. Considering that the target will have scale changes in the actual scene, it is necessary to add length and width attributes to the traditional bbox four-dimensional representation method to describe the particle properties in more detail. Create motion models for each particle separately. Where [x, y] is the coordinate of the center point of the bbox bounding box. is the movement speed of the bbox bounding box, [h, w] is the length and width of the bbox bounding box;
[0051] S102, maintain the motion state of each particle, and update the particle motion state when the next a posteriori knowledge arrives. The particle kinematic model is expressed as: In the formula, x k represents the state of the system at time k, y k represents the observed value of the system at time k, f k and h k They represent the state transfer function and system measurement function of the system, u k and v k They represent the system noise and observation noise of the system at time k respectively;
[0052] S103, based on the observed values and motion states from time 1 to time k-1, predict the motion state of the system at time k, that is, from p(x k-1 |y 1：k-1 ) to obtain p(x k |y k-1 ), given x k-1 , according to the system-related assumptions, the state x k With y k-1 Independent of each other, we can get: x for the upper formula k-1 Integrate and finally get the corresponding credibility expression: p(x k |y k-1 )=∫p(x k |x k-1 )p(x k-1 |yk-1 )dx k-1 ;
[0053] S104, update the observed value of the system at time k, that is, update p(x k |y k-1 ) to get p(x k |y k ) process, the predicted value x at time k is known k , we can get the posterior probability: Where p(y k |y k-1 ) can be taken as a normalized constant in actual calculations. The specific calculation process is: p(y k |y k-1 )=∫p(y k |x k )p(x k |Y k-1 )dx k ;
[0054] S105. In order to better extract the target appearance features, a deep appearance feature extraction network for particle filtering is constructed through a deep neural network structure of a deep over-parameterized convolutional layer (DO-Conv). The residual block of the DO-Conv convolutional layer is used as follows: Figure 3 As shown in Figure 2, the appearance features extracted by the deep appearance feature extraction network are used to calculate the weights of the particles and affect the subsequent importance ranking.
[0055] The feature extraction network used in the present invention has a specific network structure as follows: Figure 4 As shown in the figure, the input image size of the network is set to 128×64, and a 128-dimensional feature vector is finally obtained through the setting of subsequent network layer structures.
[0056] S106. Calculate the similarity d between particle i and input observation i Specifically, based on the reparameterization of the weight calculation of deep appearance features, during the importance sampling process of the particle filter, the particle filter system will calculate the similarity between the currently maintained particle attributes and the input observations of the system to obtain the weight w of the corresponding particle. i In order to add the influence of deep features to parameter calculation, the DO-Conv network model is used as the basic structure. In the importance sampling step, the model needs to first sample the original video frame according to the motion state of the particle itself, and then hand over the extracted feature vector to the weight calculation module for similarity calculation, and finally calculate and update the particle weights through the normalization method. According to the formula Get the similarity d between particle i and the input observation i , where is the deep appearance feature vector of the input observation, r i is the feature vector corresponding to a particle in the particle filter system at the current moment. The final d i That is, the similarity calculated between particle i and the input observation value. The similarity can be used as the main basis for subsequent weight calculation.
[0057] In this embodiment, step S20 specifically includes:
[0058] S201, according to the similarity d i Determine whether the observed value of the particle is reasonable. The method for determining whether the observed value is reasonable is to calculate the judgment index b. The judgment index b is calculated according to the following formula: When b is greater than the threshold t2, the current prediction value is considered reasonable.
[0059] S202, if the observed value of the particle is unreasonable, the predicted value of the particle is used to transfer the motion state, thereby generating a new particle around the original position as an adaptive result for the next step, until the next observed value arrives or the particle filter is deleted by the system. Therefore, the use of particle filtering combined with the output of the target registration module as the observed value can realize the function of target trajectory state maintenance. When the observed value is considered unreasonable, the state transfer of the motion state using the predicted value of the particle includes: Where x k represents the motion state of the particle, w k Represents the normalized weight of the particle.
[0060] Step S20 also includes S203, setting a target state maintenance model, adding auxiliary variables to each target to record the existence of the target, and the target state maintenance model is: t =[( x 1 ，t , e 1，t ),…,(x j，t ), e j,t ], j∈(1, 2, 3, …, M t ), where M t is the number of states of the tracking system at time t; x j，t represents the state of the jth target at time t; e j，t Represents x j，t The state of existence at time t; when e j，t =1, indicating x j，t Existence; when e j，t =0, indicating x j，t does not exist; the number of targets at time t can be obtained by Calculated.
[0061] Step S203 also includes: using Y 1：t ={Y1, ..., Y t} represents all observations of the tracking system at time t and before, where Y k ={x i，k |k=1,…,t,t=1,…,N k}, where Y k is the i-th observation value at the k-th time, N k is the number of observations at the kth moment, given the known observation value Y 1： At t, the state of each target can be inferred based on particle filtering.
[0062] In step S30, for a new target such as Figure 6 As shown, creating a particle filter for describing a new target includes: determining whether the unmatched target is continuously in T new If the match is successful in the frame video, it is determined to be a new target, and a particle filter is created to describe the new target.
[0063] For new targets, such as Figure 5 As shown in D2, the adjacent time state corresponding to D2 is (e D2，t+i =1,e D2，t =1), at this time, the target is considered to be a new target to be confirmed, and the adjacent time states of targets D3 and D4 are respectively (e D3，t+j =1,e D3，t+1 =1) and (e D4，t+k =1,e D4，t+j =1), at time t+j, since D2 has been continuously new The match is successful in the frame video, so the target is confirmed to be a new target and can be used as a normal matching target later.
[0064] In step S30, for the disappeared target Figure 7 As shown, deleting the existing particle filter describing the disappeared target includes: determining whether the unmatched target is continuously in T dead If the matching fails in the frame video, the target is determined to have disappeared, and the particle filter used to describe the disappeared target is deleted.
[0065] For the disappearing target, Figure 5 As shown in A1, A2 and A3 in , each target that has been attempted to be matched will correspond to a certain state at each moment, and the state will change according to the input of the next sequence. Figure 5 The state of A1 at the adjacent time is (e A1，t =0,e A1，t-1=0), at this time, the target is considered to be a target to be confirmed as disappeared, and the states of targets A2 and A3 at adjacent moments are respectively (e A2，t+i =0,e A2，t =0) and (e A3，t+i =0,e A3，t =0), at this time, since the target A1 has been continuously in T dead The matching fails in the frame video, so the target is confirmed to be a disappeared target and the deletion operation is performed.
[0066] As a preferred embodiment, after step S10, the method further includes step S11, performing an improved particle filtering method based on an over-parameterized feature network for the successfully matched target, updating the particle filter state of the successfully matched target, updating the target trajectory, and entering S40.
[0067] The multi-target tracking method of the present invention, in a multi-target tracking scenario, based on an over-parameterized feature network and particle filtering, can adapt to complex problems such as dynamic environments, apparent changes, and changes in the number of targets in real scenarios, and achieve effective and stable multi-target tracking results. The multi-target tracking method of the present invention performs model design and system construction based on the multi-target tracking technology of detection, uses a deep neural network to extract target features, performs similarity calculation on the corresponding target depth features, and uses a variety of related methods such as the center loss function to optimize the feature extraction network. In addition, a multi-level matching method is used to improve the accuracy of target matching, thereby obtaining a more effective registration model. The multi-target tracking method of the present invention can significantly improve the registration accuracy of detection results and existing targets in the process of target tracking and trajectory generation in complex scenarios, and has important application value in the fields of passenger flow analysis, behavior modeling, and unmanned driving.
[0068] The multi-target detection method of the present invention, after being processed by the detection and target registration module, the multi-target tracking model will match the detection result in the current frame with the previously confirmed target trajectory. However, only the matching result cannot form a continuous tracking trajectory, nor can it determine the new or disappearing situation of the target. Therefore, the tracking model requires a trajectory generation module to perform persistent processing on the registration result, mark the position of the same target in different frames, and finally obtain the tracking trajectory corresponding to a certain target. The random number scheme of the present invention adopts particle filtering as the main method of the trajectory generation module. However, the result after registration may still have certain mismatches or missing matches. If the registration result is directly used as the tracking trajectory, it may cause deviations in the tracking trajectory and have a negative impact on the performance of the tracking model after long-term error accumulation.
[0069] Therefore, the trajectory generation module needs to optimize and correct the alignment results while maintaining the tracking trajectory, using its own advantages to obtain a more stable tracking trajectory. In addition, in the multi-target tracking scenario, there is also the problem of changes in the number of targets, that is, the birth and disappearance of targets. Some targets will enter or leave the field of view over time, resulting in the same target having multiple states in the tracking task, and the states will be transformed into each other. The trajectory generation module needs to have a certain ability to handle new or disappeared targets. In order to realize the trajectory generation module and at the same time ensure the stability of its trajectory and the management of new or disappeared targets, the present invention introduces deep features to improve the weight reliability of the particle filter; the state management of the target is performed in the particle filter to realize the judgment of new and disappeared targets, thereby realizing the design and optimization of the trajectory generation module in the multi-target tracking framework.
[0070] The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the multi-target tracking method as described above is implemented.
[0071] The present invention also provides an electronic device, including a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the multi-target tracking method as described above is implemented.
[0072] Although the content of the present invention has been described in detail through the above preferred embodiments, it should be appreciated that the above description should not be considered as a limitation of the present invention. After reading the above content, it will be apparent to those skilled in the art that various modifications and substitutions of the present invention will occur. Therefore, the protection scope of the present invention should be limited by the appended claims.
Claims
1. A multi-target tracking method, characterized in that: The multi-target tracking method comprises: S10, obtaining picture information of the current frame, extracting target information to be matched of the current frame, matching the target information to be matched with existing targets extracted from previous frames, and obtaining successfully matched targets and unmatched targets; S20, for the unmatched target, determining whether it is a new target or a disappeared target; S30, for a new target, creating a particle filter for describing the new target; for a disappeared target, deleting an existing particle filter for describing the disappeared target; S40: Output trajectory tracking results.
2. The multi-target tracking method according to claim 1, characterized in that: Step S10 includes: S101. Create motion models for each particle Where [x, y] is the coordinate of the center point of the bbox bounding box. is the movement speed of the bbox bounding box, [h, w] is the length and width of the bbox bounding box; S102, maintaining the motion state of each particle, and updating the particle motion state when the next a posteriori knowledge arrives. The kinematic model of the particle is expressed as: In the formula, x k represents the state of the system at time k, y k represents the observed value of the system at time k, f k and h k They represent the state transfer function and system measurement function of the system, u k and k They represent the system noise and observation noise of the system at time k respectively; S103, based on the observed values and motion states from time 1 to k-1, predict the motion state of the system at time k, that is, from p(x k-1 |y 1：k-1 ) to obtain p(x k |y k-1 ), given x k-1 , according to the system-related assumptions, the state x k With y k-1 Independent of each other, we can get: x for the upper formula k-1 Integrate and finally get the corresponding credibility expression: p(x k |y k-1 )=∫p(x k |x k-1 )p(x k-1 |y k-1 )dx k-1 ; S104, update the observed value of the system at time k, that is, update p(x k |y k-1 ) to get p(x k |y k ) process, the predicted value x at time k is known k , we can get the posterior probability: Where p(y k |y k-1 ) can be taken as a normalized constant in actual calculations. The specific calculation process is: p(y k |y k-1 )=∫p(y k |x k )p(x k |Y k-1 )dx k ; S105, constructing a deep appearance feature extraction network for particle filtering through a deep neural network structure of a depthwise over-parameterized convolutional layer (DO-Conv), and using the appearance features extracted by the deep appearance feature extraction network to calculate the weights of the particles and affect the subsequent importance ranking; S106. Calculate the similarity d between particle i and input observation i .
3. The multi-target tracking method according to claim 2, characterized in that: The input image size of the deep appearance feature extraction network is 128×64, and a 128-dimensional feature vector is finally obtained by setting the network layer structure.
4. The multi-target tracking method according to claim 2, characterized in that: In S106, according to the formula Get the similarity d between particle i and the input observation i , where is the deep appearance feature vector of the input observation, r i It is the feature vector corresponding to a particle in the particle filter system at the current moment.
5. The multi-target tracking method according to claim 2, characterized in that: Step S20 specifically: S201, according to the similarity d i Determine whether the observed value of the particle is reasonable; S202. If the observed value of the particle is unreasonable, the predicted value of the particle is used to transfer the motion state.
6. The multi-target tracking method according to claim 5, characterized in that: Step S20 also includes S203, setting a target state maintenance model, adding auxiliary variables to each target to record the existence of the target, and the target state maintenance model is: t =[(x 1，t , e 1，t ),…,(x j，t ), e j,t ], j∈(1, 2, 3, …, Mt), where M t is the number of states of the tracking system at time t; x j，t represents the state of the jth target at time t; e j，t Represents x j，t The state of existence at time t; when e j，t =1, indicating x j，t Existence; when e j，t =0, indicating x j，t does not exist; the number of targets at time t can be obtained by Calculated.
7. The multi-target tracking method according to claim 6, characterized in that: Step S203 also includes: using Y 1：t ={Y1, ..., Y t } represents all observations of the tracking system at time t and before, where Y k ={x i，k |k=1,…,t,t=1,…,N k }, where Y k is the i-th observation value at the k-th time, N k is the number of observations at the kth moment, given the known observation value Y 1：t When , the state of each target can be inferred based on the particle filter.
8. The multi-target tracking method according to claim 5, characterized in that: In step S201, the judgment index b is calculated, and the judgment index b is calculated according to the following formula: When b is greater than the threshold t2, the current prediction value is considered reasonable.
9. The multi-target tracking method according to claim 5, characterized in that: In step S202, using the predicted value of the particle to transfer the motion state includes: Where x k represents the motion state of the particle, w k Represents the normalized weight of the particle.
10. The multi-target tracking method according to claim 1, characterized in that: In step S30, for a new target, creating a particle filter for describing the new target includes: determining whether the unmatched target is continuously in T new If the match is successful in the frame video, it is determined to be a new target, and a particle filter for describing the new target is created.
11. The multi-target tracking method according to claim 1, characterized in that: In step S30, for a disappeared target, deleting the existing particle filter describing the disappeared target includes: determining whether the unmatched target is continuously in T dead If the matching fails in the frame video, the target is determined to have disappeared, and the particle filter used to describe the disappeared target is deleted.
12. The multi-target tracking method according to claim 1, characterized in that: After step S10, the process further includes step S11: for a successfully matched target, updating the particle filter state of the successfully matched target, and then proceeding to S40.
13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the multi-target tracking method according to any one of claims 1 to 12 is implemented.
14. An electronic device, characterized in that: The method comprises a processor and a memory, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the multi-target tracking method according to any one of claims 1 to 12 is implemented.
Citation Information
Cited By
Multi-target tracking method, storage medium, and electronic device
WO2026138121A1