A target positioning method and device applied to a multi-agent system

Abstract

This invention relates to a target localization method and apparatus for multi-agent systems, belonging to the field of multi-agent cooperative control technology, and solves the problem of inconsistent target localization caused by the attitude differences and transmission delays of various agents in existing multi-agent target detection technologies. The method includes: synchronizing the time and coordinate system of the first target point clouds of each agent to obtain the second target point clouds of each agent; matching the second target point clouds based on distance similarity and edge similarity to obtain the comprehensive matching degree of each second target point cloud; filtering the second target point clouds according to the comprehensive matching degree to obtain the third target point cloud; calculating the comprehensive weight coefficient of the third target point cloud based on the target observation angle and target observation distance of the agent corresponding to the third target point cloud; and calculating the target position using a weighted summation method based on each third target point cloud and its comprehensive weight coefficient.

Description

Technical Field

[0001] This invention relates to the field of multi-agent cooperative control technology, and in particular to a target localization method and apparatus for multi-agent systems. Background Technology

[0002] Target detection is a core perception task for intelligent agents such as vehicles and drones, and its accuracy directly affects the agent's decision-making and driving safety. Detection by a single intelligent agent has inherent shortcomings such as occlusion and missed detection of distant targets. Therefore, multi-agent collaborative perception has become an important development direction. By exchanging information between intelligent agents and integrating complementary perception data, the comprehensiveness and accuracy of detection can be improved.

[0003] Current multi-agent cooperative target detection technologies generally assume ideal perception environment conditions, meaning that the agents are assumed to have completely consistent postures (positions, angles) and no time delay in information transmission. However, in actual deployments, errors exist in the localization modules of each agent, leading to deviations in the position labeling of the same target in different agent coordinate systems, causing feature space misalignment. Simultaneously, communication between agents is susceptible to congestion and interruptions, resulting in asynchronous timestamps of transmitted information, further exacerbating the feature fusion misalignment problem. Summary of the Invention

[0004] Based on the above analysis, the embodiments of the present invention aim to solve the problem of inconsistent target localization caused by the pose differences and transmission delays of each agent in the existing multi-agent target detection technology.

[0005] In a first aspect, embodiments of the present invention provide a target localization method applied to a multi-agent system, comprising: Synchronize the first target point cloud of each intelligent agent with time and coordinate system one to obtain the second target point cloud of each intelligent agent; The second target point cloud is matched based on distance similarity and edge similarity to obtain the comprehensive matching degree of each second target point cloud; The second target point cloud is filtered based on the comprehensive matching degree to obtain the third target point cloud; Based on the target observation angle and target observation distance of the agent corresponding to the third target point cloud, calculate the comprehensive weight coefficient of the third target point cloud; and The position of the target is calculated by weighted summation based on each of the third target point clouds and their comprehensive weight coefficients.

[0006] Based on the further improvement of the target localization method described above, before performing time synchronization and coordinate system unification on the first target point cloud collected by each agent, the following steps are also included: The original target point cloud collected by each intelligent agent is preprocessed to obtain the first target point cloud of each intelligent agent; wherein, the preprocessing includes noise reduction processing and downsampling processing.

[0007] Based on the further improvement of the above target localization method, each of the second target point clouds is matched based on distance similarity and edge similarity to obtain the comprehensive matching degree of each of the second target point clouds, including: Based on the Hausdorff distance between each of the second target point clouds, calculate the distance similarity between each of the second target point clouds; The edge similarity between each second target point cloud is calculated based on the cosine similarity between the edge features of each second target point cloud. Based on the distance similarity and edge similarity between each of the second target point clouds, calculate the comprehensive similarity between each of the second target point clouds; and The overall similarity between each second target point cloud and other second target point clouds is summed to obtain the overall matching degree of each second target point cloud.

[0008] Based on further improvements to the above target localization method, the second target point cloud is filtered according to the comprehensive matching degree to obtain a third target point cloud, including: From each of the second target point clouds, select N second target point clouds whose comprehensive matching degree is greater than or equal to the comprehensive matching degree threshold, where N is a positive integer; If N is greater than or equal to K, then select the top K second target point clouds with the highest comprehensive matching degree from the N second target point clouds, where K is a preset positive integer; If N is less than K, the overall matching degree threshold is gradually reduced until N is greater than or equal to K.

[0009] Based on the further improvement of the target localization method described above, the comprehensive weighting coefficient of the third target point cloud is calculated according to the target observation angle and target observation distance of the agent corresponding to the third target point cloud, including: Calculate the angle weighting coefficient of each third target point cloud based on the target observation angle of the agent corresponding to each third target point cloud; Based on the target observation distance of the agent corresponding to each of the third target point clouds, calculate the distance weighting coefficient of each of the third target point clouds; and The angle weighting coefficient and the distance weighting coefficient of each of the third target point clouds are normalized and weighted and fused to obtain the comprehensive weighting coefficient of each of the third target point clouds.

[0010] Based on the further improvement of the above target localization method, the angle weighting coefficient The calculation formula is:

[0011] Where β is the adjustment coefficient. Let be the observation angle of the target in the local coordinate system of the agent corresponding to the j-th third target point cloud. This refers to the angle weighting coefficient of the j-th third target point cloud.

[0012] Based on further improvements to the above target localization method, the distance weighting coefficient The calculation formula is:

[0013] Where γ is the attenuation coefficient. Let be the observation distance of the target in the local coordinate system of the agent corresponding to the j-th third target point cloud. This refers to the distance weighting coefficient of the j-th third target point cloud.

[0014] Based on further improvements to the above target localization method, the comprehensive weighting coefficient... The calculation formula is:

[0015] Where K is the number of points in the third target cloud. is the comprehensive weight coefficient of the j-th third target point cloud.

[0016] Based on the further improvement of the above target localization method, the target position is calculated by weighted summation according to each of the third target point clouds and their comprehensive weight coefficients, including: Calculate the center coordinates of each of the third target point clouds. , , ), j∈1,2,3…,K, where K is the number of points in the third target cloud; Calculate the target's location using the following formula ( ):

[0017] in, is the comprehensive weight coefficient of the j-th third target point cloud.

[0018] Secondly, an embodiment of the present invention provides a target localization device applied to a multi-agent system, comprising: The synchronization unit is used to synchronize the first target point cloud of each intelligent agent with time and coordinate system one, so as to obtain the second target point cloud of each intelligent agent. A matching unit is used to match each of the second target point clouds based on distance similarity and edge similarity to obtain the comprehensive matching degree of each of the second target point clouds; A filtering unit is used to filter the second target point cloud according to the comprehensive matching degree to obtain a third target point cloud; The first calculation unit is used to calculate the comprehensive weight coefficient of the third target point cloud based on the target observation angle and target observation distance of the agent corresponding to the third target point cloud; and The second calculation unit is used to calculate the position of the target by weighted summation based on each of the third target point clouds and their comprehensive weight coefficients.

[0019] Compared with the prior art, the present invention can achieve at least one of the following beneficial effects: 1. This solution overcomes the limitations of single agent perspective and detection range by adopting a multi-agent collaborative observation architecture; it achieves accurate clustering of multi-source point clouds by using two-dimensional feature matching of "edge similarity + distance similarity"; and then outputs a unified target localization result in the global coordinate system through an adaptive weighted fusion algorithm, which solves the problems of single agent localization deviation and inconsistency of multi-agent localization results, and improves the overall accuracy of target localization.

[0020] 2. For high-confidence target point clouds from different agents, a two-factor weighting mechanism based on observation angle and observation distance was designed. That is, by giving higher weight to high-quality data from frontal and close-range observations, and reducing the weight of degraded observation data from large angles and long distances, the negative impact of low-quality observation results on the final localization is avoided. The weight design makes the target position of the fused output closer to the true value.

[0021] In this invention, the above-described technical solutions can be combined with each other to achieve more preferred combinations. Other features and advantages of this invention will be set forth in the following description, and some advantages may become apparent from the description or be learned by practicing the invention. The objects and other advantages of this invention can be realized and obtained from what is particularly pointed out in the description and drawings. Attached Figure Description

[0022] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Figure 1 A flowchart illustrating a target localization method for a multi-agent system according to an embodiment of the present invention is shown.

[0023] Figure 2 A structural block diagram of a target localization device applied to a multi-agent system according to an embodiment of the present invention is shown. Detailed Implementation

[0024] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0025] According to one aspect of the present invention, a target localization method for use in a multi-agent system is provided. Figure 1 A flowchart illustrating a target localization method for a multi-agent system according to an embodiment of the present invention is shown. Figure 1 As shown, the target localization method applied to a multi-agent system includes: Step 101: Synchronize the time and coordinate system of the first target point cloud of each agent to obtain the second target point cloud of each agent.

[0026] The intelligent agent is equipped with devices such as LiDAR or depth camera point cloud to acquire point cloud data.

[0027] In this embodiment, the first target point cloud of the agent can be the target point cloud obtained after preprocessing the original target point cloud collected by the agent.

[0028] In some embodiments, preprocessing of the target raw point cloud includes noise reduction and downsampling.

[0029] For noise reduction, a statistical filtering algorithm can be used to remove isolated noise from the point cloud. Specifically, the number of neighboring points can be set to L. Then, for each point in the target original point cloud, L nearest neighbors are searched and the Euclidean distance is calculated. When the distance between a point in the target original point cloud and its other neighbors is greater than μ + 2σ, the point is determined to be a noise point and deleted, where μ is the average distance between points in the neighboring area and σ is the standard deviation of the distance between points in the neighboring area.

[0030] For downsampling, voxel lattice filtering can be used to reduce the amount of point cloud data and improve the efficiency of subsequent processing based on a preset voxel size.

[0031] In this embodiment, the high-precision system time of the control center can be used. This serves as a global baseline time anchor. Each agent synchronously records its local timestamp when acquiring the target's original point cloud. The original target point cloud and local timestamp are transmitted to the control center via a communication network. The control center calculates the deviation between the local timestamp of each agent and the global reference time anchor. If the local timestamp is inconsistent with the global reference time anchor, the first target point cloud with the smallest deviation from the global reference time anchor is selected from the first target point cloud collected by the agent at a timestamp adjacent to the global reference time anchor as the first target point cloud after time synchronization.

[0032] In this embodiment, the first target point cloud of each agent after time synchronization can be transformed from the agent's own coordinate system to the global coordinate system. The transformation formula is as follows:

[0033] Among them, X agent R represents the coordinates of the first target point cloud after time synchronization in the corresponding agent's own coordinate system, where R is the rotation matrix and T is the position of the agent in the global coordinate system (which can be obtained through the agent's own localization system). This is the first target point cloud after coordinate system one.

[0034] In this embodiment, the second target point cloud is the first target point cloud that has simultaneously achieved time synchronization and coordinate system one.

[0035] Step 102: Match each of the second target point clouds based on distance similarity and edge similarity to obtain the comprehensive matching degree of each of the second target point clouds.

[0036] In this embodiment, the improved Hausdorff distance can be used to calculate the distance similarity between two second target point clouds, and the specific formula is as follows:

[0037] in, Let be the distance similarity between the i-th second target point cloud and the j-th second target point cloud. and These are the i-th and j-th second target point clouds, respectively. and These are the second target point clouds. and The number of point clouds, It is Euclidean distance.

[0038] In this embodiment, the edge similarity between the edge features of each second target point cloud can be calculated based on the cosine similarity between them. In this embodiment, edge features of two second target point clouds can be extracted first. Specifically, the second target point clouds can be clustered to obtain estimated target coordinate positions; then, edge features are constructed by connecting the estimated target coordinate positions to the origin of the second target point clouds, which are unit vectors pointing from the origin to the estimated target coordinate positions. When there is only a single target in the second target point cloud, the edge features contain only a single vector; when there are multiple targets in the second target point cloud, the edge features contain the same number of unit vectors as the number of targets.

[0039] In this embodiment, the edge similarity between two second target point clouds can be calculated using the following formula:

[0040] in, For edge similarity, For the edge features of the i-th second target point cloud, The edge features of the j-th second target point cloud are given.

[0041] In this embodiment, the comprehensive similarity between the various second target point clouds can be calculated based on the distance similarity and edge similarity between them. Specifically, the distance similarity between the various second target point clouds can first be normalized as follows:

[0042] in, Let be the distance similarity between the i-th second target point cloud and the j-th second target point cloud. The minimum value of all distance similarities. This represents the maximum value of all distance similarities.

[0043] Then, a weighted summation method can be used to fuse distance similarity and edge similarity to obtain a comprehensive similarity. The formula is as follows:

[0044] in, Let α be the comprehensive similarity between the i-th second target point cloud and the j-th second target point cloud, and let α be the weighting coefficient.

[0045] In this embodiment, the overall similarity between each second target point cloud and other second target point clouds can be summed to obtain the overall matching degree of each second target point cloud. For example, if there are three second target point clouds (P1, P2, P3), and the overall similarity between the first second target point cloud P1 and the other two second target point clouds P2 and P3 is S12 and S13 respectively, then the overall matching degree of the first second target point cloud P1 is S12 plus S13.

[0046] Step 103: Filter the second target point cloud according to the comprehensive matching degree to obtain the third target point cloud.

[0047] In this embodiment, N second target point clouds with a comprehensive matching degree greater than or equal to a comprehensive matching degree threshold can be selected from each second target point cloud, where N is a positive integer. The second target point clouds with a comprehensive matching degree greater than or equal to the comprehensive matching degree threshold are the second target point clouds with high confidence.

[0048] In this embodiment, if the number N of high-confidence second target point clouds is greater than or equal to K, then the top K second target point clouds with the highest comprehensive matching degree are selected from the N high-confidence second target point clouds, where K is a preset positive integer. In this embodiment, to ensure the accuracy of the target localization result, K can be set to multiple values ​​(e.g., 3).

[0049] In this embodiment, if N is less than K, the overall matching threshold can be gradually reduced until K or more high-confidence second target point clouds are selected.

[0050] In this embodiment, the K high-confidence second target point clouds are the K third target point clouds.

[0051] Step 104: Calculate the comprehensive weighting coefficient of the third target point cloud based on the target observation angle and target observation distance of the agent corresponding to the third target point cloud.

[0052] In this embodiment, for each third target point cloud, the angle weighting coefficient and the distance weighting coefficient can be calculated according to the target observation angle and target observation distance of its corresponding agent.

[0053] In this embodiment, the angle weighting coefficient can be used to measure the observation angle of the agent towards the target. The smaller the observation angle, that is, the closer the agent is to the frontal observation target, the more accurate the observation information it obtains, and the greater the weight.

[0054] In some embodiments, angle weighting coefficient The calculation formula is:

[0055] Where β is the adjustment coefficient. Let be the observation angle of the target in the local coordinate system of the agent corresponding to the j-th third target point cloud. This refers to the angle weighting coefficient of the j-th third target point cloud.

[0056] In this embodiment, the distance weighting coefficient can be used to measure the observation distance of the agent to the target. The smaller the observation distance, that is, the closer the agent is to the target, the more accurate the observation information it obtains, and the greater the weight.

[0057] In some embodiments, distance weighting coefficient The calculation formula is:

[0058] Where γ is the attenuation coefficient. Let be the observation distance of the target in the local coordinate system of the agent corresponding to the j-th third target point cloud. This refers to the distance weighting coefficient of the j-th third target point cloud.

[0059] In this embodiment, the angle weighting coefficient and the distance weighting coefficient can be normalized and weighted to obtain the comprehensive weighting coefficient. The specific formula is as follows: Comprehensive weighting coefficient The calculation formula is:

[0060] Where K is the number of points in the third target cloud. is the comprehensive weight coefficient of the j-th third target point cloud.

[0061] Step 105: Calculate the position of the target by weighted summation based on each of the third target point clouds and their comprehensive weight coefficients.

[0062] In this embodiment, the center coordinates of each third target point cloud can be calculated first. , , The calculation formula is:

[0063] in, For the third target point cloud The number of points, For the third target point cloud The global coordinates of the interior point.

[0064] Then, the third target point cloud can be calculated according to the following formula. The location of the target ( ):

[0065] in, is the comprehensive weight coefficient of the j-th third target point cloud.

[0066] In this embodiment, the optimal integration of multi-source high-confidence observation information is achieved by weighted summation, which suppresses the cumulative effect of single-source observation errors and outputs a consistent target positioning estimate in the global coordinate system, ensuring the high accuracy and robustness of the fusion result.

[0067] According to another aspect of the present invention, a target localization device for use in a multi-agent system is provided. For example... Figure 2As shown, the device 200 includes: a synchronization unit 201, used to synchronize the time and coordinate system of the first target point clouds of each agent to obtain the second target point clouds of each agent; a matching unit 202, used to match each of the second target point clouds based on distance similarity and edge similarity to obtain the comprehensive matching degree of each of the second target point clouds; a filtering unit 203, used to filter the second target point clouds according to the comprehensive matching degree to obtain the third target point cloud; a first calculation unit 204, used to calculate the comprehensive weight coefficient of the third target point cloud according to the target observation angle and target observation distance of the agent corresponding to the third target point cloud; and a second calculation unit 205, used to calculate the position of the target by weighted summation according to each of the third target point clouds and its comprehensive weight coefficient.

[0068] It is understandable that unit 201 in device 200 Operation of Unit 205 Figure 1 Step 101 Step 105 is similar and will not be described in detail here.

[0069] Compared with the prior art, the embodiments of the present invention can achieve at least one of the following beneficial effects: 1. This solution overcomes the limitations of single agent perspective and detection range by adopting a multi-agent collaborative observation architecture; it achieves accurate clustering of multi-source point clouds by using two-dimensional feature matching of "edge similarity + distance similarity"; and then outputs a unified target localization result in the global coordinate system through an adaptive weighted fusion algorithm, which solves the problems of single agent localization deviation and inconsistency of multi-agent localization results, and improves the overall accuracy of target localization.

[0070] 2. For high-confidence target point clouds from different agents, a two-factor weighting mechanism based on observation angle and observation distance was designed. That is, by giving higher weight to high-quality data from frontal and close-range observations, and reducing the weight of degraded observation data from large angles and long distances, the negative impact of low-quality observation results on the final localization is avoided. The weight design makes the target position of the fused output closer to the true value.

[0071] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A target localization method applied to a multi-agent system, characterized in that, include: Synchronize the first target point cloud of each intelligent agent with time and coordinate system one to obtain the second target point cloud of each intelligent agent; The second target point cloud is matched based on distance similarity and edge similarity to obtain the comprehensive matching degree of each second target point cloud; The second target point cloud is filtered based on the comprehensive matching degree to obtain the third target point cloud; Calculate the comprehensive weighting coefficient of the third target point cloud based on the target observation angle and target observation distance of the agent corresponding to the third target point cloud; as well as The position of the target is calculated by weighted summation based on each of the third target point clouds and their comprehensive weight coefficients.

2. The target localization method according to claim 1, characterized in that, Before synchronizing the time and coordinate system of the first target point cloud collected by each intelligent agent, the following steps are also included: The original target point cloud collected by each intelligent agent is preprocessed to obtain the first target point cloud of each intelligent agent; wherein, the preprocessing includes noise reduction processing and downsampling processing.

3. The target localization method according to claim 1, characterized in that, The second target point cloud is matched based on distance similarity and edge similarity to obtain the comprehensive matching degree of each second target point cloud, including: Based on the Hausdorff distance between each of the second target point clouds, calculate the distance similarity between each of the second target point clouds; The edge similarity between each second target point cloud is calculated based on the cosine similarity between the edge features of each second target point cloud. Based on the distance similarity and edge similarity between each of the second target point clouds, calculate the comprehensive similarity between each of the second target point clouds; and The overall similarity between each second target point cloud and other second target point clouds is summed to obtain the overall matching degree of each second target point cloud.

4. The target localization method according to claim 1, characterized in that, The second target point cloud is filtered based on the comprehensive matching degree to obtain the third target point cloud, including: From each of the second target point clouds, select N second target point clouds whose comprehensive matching degree is greater than or equal to the comprehensive matching degree threshold, where N is a positive integer; If N is greater than or equal to K, then select the top K second target point clouds with the highest comprehensive matching degree from the N second target point clouds, where K is a preset positive integer; If N is less than K, the overall matching degree threshold is gradually reduced until N is greater than or equal to K.

5. The target localization method according to claim 1, characterized in that, Based on the target observation angle and target observation distance of the agent corresponding to the third target point cloud, the comprehensive weighting coefficient of the third target point cloud is calculated as follows: Calculate the angle weighting coefficient of each third target point cloud based on the target observation angle of the agent corresponding to each third target point cloud; Based on the target observation distance of the agent corresponding to each of the third target point clouds, calculate the distance weighting coefficient of each of the third target point clouds; and The angle weighting coefficient and the distance weighting coefficient of each of the third target point clouds are normalized and weighted and fused to obtain the comprehensive weighting coefficient of each of the third target point clouds.

6. The target localization method according to claim 5, characterized in that, The angle weighting coefficient The calculation formula is: Where β is the adjustment coefficient. Let be the observation angle of the target in the local coordinate system of the agent corresponding to the j-th third target point cloud. This refers to the angle weighting coefficient of the j-th third target point cloud.

7. The target localization method according to claim 6, characterized in that, The distance weighting coefficient The calculation formula is: Where γ is the attenuation coefficient. Let be the observation distance of the target in the local coordinate system of the agent corresponding to the j-th third target point cloud. This refers to the distance weighting coefficient of the j-th third target point cloud.

8. The target localization method according to claim 7, characterized in that, The comprehensive weighting coefficient The calculation formula is: Where K is the number of points in the third target cloud. is the comprehensive weight coefficient of the j-th third target point cloud.

9. The target localization method according to claim 1, characterized in that, The position of the target is calculated using a weighted summation method based on each of the third target point clouds and its comprehensive weight coefficients, including: Calculate the center coordinates of each of the third target point clouds. , , ), j∈1,2,3…,K, where K is the number of points in the third target cloud; Calculate the target's location using the following formula ( ): in, is the comprehensive weight coefficient of the j-th third target point cloud.

10. A target localization device for use in a multi-agent system, characterized in that, include: The synchronization unit is used to synchronize the first target point cloud of each intelligent agent with time and coordinate system one, so as to obtain the second target point cloud of each intelligent agent. A matching unit is used to match each of the second target point clouds based on distance similarity and edge similarity to obtain the comprehensive matching degree of each of the second target point clouds; A filtering unit is used to filter the second target point cloud according to the comprehensive matching degree to obtain a third target point cloud; The first calculation unit is used to calculate the comprehensive weight coefficient of the third target point cloud based on the target observation angle and target observation distance of the agent corresponding to the third target point cloud. as well as The second calculation unit is used to calculate the position of the target by weighted summation based on each of the third target point clouds and their comprehensive weight coefficients.

Images