A multi-target fusion tracking method and device based on a perception area

Abstract

The application discloses a kind of multi-target fusion tracking method and device based on perception area, belong to automatic driving technical field, first design environment complexity judging mechanism, according to multiple information in traffic scene, judge environment classification;Again according to the type of environment, select data association algorithm to realize fast data association;Then according to the motion state of ego vehicle and target, and the current data association state establishes dynamic parameter threshold system, according to the threshold parameter of environment adjustment, real-time response to different working conditions.The application processes various data such as sensor information, vehicle information, road information, etc., to respond to different automatic driving scenes in the optimal scheme, which can make the tracking of target more stable and reliable.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving technology, and more specifically, relates to a multi-target fusion tracking method and apparatus based on a perception area. Background Technology

[0002] In the field of autonomous driving, the scenarios that need to be covered for the fusion of target objects are becoming increasingly complex and diverse. Currently, most solutions for target object fusion tracking focus on a single algorithm or technical means, but their robustness and universality in dealing with real-time road conditions are generally limited.

[0003] The most widely used perception fusion technology currently is the fusion and target tracking method based on monocular vision and millimeter-wave radar. The shortcomings of existing perception fusion methods can be explained from the following three aspects:

[0004] 1. For example, image recognition relying on cameras is greatly affected by the lighting conditions in the actual scene. Strong light or rainy / foggy weather can cause target recognition to fail. In such cases, it will have an incalculable negative impact on the fusion of radar and camera, causing the entire intelligent driving system to malfunction.

[0005] 2. There are shortcomings in detecting the state information of the target object, such as distance. The camera relies on pixels to calculate the distance, which is not very accurate and will affect the state information of the target during the fusion stage.

[0006] 3. Most current fusion solutions require the support of sensors such as LiDAR and cameras, which are costly. The synchronous operation of multiple sensors also places high demands on hardware, resulting in insufficient stability and complex maintenance. Summary of the Invention

[0007] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention proposes a multi-target fusion tracking method and apparatus based on a sensing area. It provides a composite fusion mechanism that selects different fusion methods and threshold parameters for real-time road conditions to perform target data fusion, thereby achieving tracking stability and better robustness.

[0008] To achieve the above objectives, according to one aspect of the present invention, a multi-target fusion tracking method based on a perception region is provided, comprising:

[0009] Based on the lane line shape collected by the sensor, the area to be sensed is divided into multiple sensing subdomains, and each sensing subdomain is classified according to the processing order.

[0010] The complexity of the environment is determined in each sensing subdomain, and the data association method used in each sensing subdomain is determined based on the complexity of the environment;

[0011] The threshold parameters in the filtering algorithm are dynamically adjusted based on the current data association method of the target within the sensing subdomain and the sensor data status.

[0012] The target object is tracked by data fusion based on the threshold parameters corresponding to each sensing subdomain, and the target state calculated by fusion is output.

[0013] In some alternative implementations, the process of dividing the area to be sensed into multiple sensing sub-domains based on the lane line shape acquired by the sensors includes:

[0014] At the current moment, lane line data is collected, and a perception area is constructed in front of the vehicle based on the shape of the lane line. The width of the perception area is the total lane width, and the longitudinal direction starts from the front axle of the vehicle and extends forward along the lane line by a first preset distance.

[0015] The sensing area is divided into multiple sensing sub-domains based on the shape of the lane lines and the longitudinal distance. The width of each sensing sub-domain is the width of one lane, and the longitudinal distance is along the lane area with a second preset length.

[0016] The processing order of the set sensing subdomains is filled into the queue to obtain the processing subdomain queue of the sensing region.

[0017] In some optional implementations, the step of determining the environmental complexity in each sensing subdomain and determining the data association method used in each sensing subdomain based on the environmental complexity includes:

[0018] For the current perceptual subdomain to be processed in the queue, calculate the complexity of the current scene in the perceptual subdomain. The more objects there are in the current scene of the perceptual subdomain, the higher the complexity of the perceptual subdomain.

[0019] When the sensing subdomain is determined to be a simple working condition, the sensing measurement data of the sensors within the sensing subdomain is obtained to construct a sensing measurement list. Historical targets that already exist within the sensing subdomain are identified. A historical target list is constructed based on the length of time each historical target has existed. Each historical target in the historical target list is matched with each sensing measurement value in the sensing measurement list to construct the correspondence between each historical target and the sensing measurement value.

[0020] When a complex working condition is determined within the sensing subdomain, the sensing measurement data of the sensors within the sensing subdomain are acquired to construct a sensing measurement list. Historical targets that already exist within the sensing subdomain are identified, and a historical target list is constructed based on the duration of existence of each historical target. For each historical target, a coarse threshold covering various dimensions is set. The Mahalanobis distance between each historical target and the sensing measurement values ​​in the sensing measurement list within the sensing subdomain that meet the respective coarse threshold range is calculated, and a Mahalanobis distance matrix is ​​constructed. Positions in the Mahalanobis distance matrix that are not within the coarse threshold range are set to 0. The upper limit of the number of sensing measurement values ​​that each historical target participates in the calculation is controlled, and a Mahalanobis distance confirmation matrix is ​​constructed.

[0021] For the Mahalanobis distance confirmation matrix, all positions that are target values ​​are confirmed. Based on the rule that one perception measurement value corresponds to one historical target, a matrix of all feasible events is generated to obtain a cluster of feasible events. The feasible events are presented in the form of a matrix, and each row and column can only contain one target value.

[0022] Calculate the association probability of each event in the feasible event cluster, statistically calculate the joint probability of each historical target, and obtain the joint probability matrix.

[0023] In some alternative implementations, by Calculate the complexity result within the current scene of the perceptual subdomain. n is the number of target objects in the current scene of the perception subdomain, T constant The threshold is complex. result=0 indicates that the sensing subdomain is in a simple working condition, and result=1 indicates that the sensing subdomain is in a complex working condition.

[0024] In some optional implementations, the step of matching each historical target in the historical target list with each perception measurement value in the perception measurement list to construct a correspondence between each historical target and perception measurement value includes:

[0025] Calculate the Mahalanobis distance between each historical target and each perceived measurement;

[0026] Based on the target perception measurement value that matches the previous frame of the historical target being processed, the Mahalanobis distance value between the historical target being processed at the current moment and the target perception measurement value is corrected by the number of historical matches.

[0027] Following the order of historical targets in the historical target list, the perceptual measurement values ​​corresponding to the minimum Mahalanobis distance between the historical targets are selected sequentially from the perceptual measurement list to construct the correspondence between the historical targets and the perceptual measurement values ​​corresponding to the minimum Mahalanobis distance.

[0028] In some optional implementations, the step of correcting the Mahalanobis distance value between the historical target being processed at the current moment and the target perception measurement value based on the target perception measurement value matched with the previous frame of the historical target being processed, by the number of historical matches, includes:

[0029] Each initial input sensory measurement is assigned an ID, and the distance and velocity variation ranges for each ID are defined. If an input sensory measurement in the next frame falls within the range of a given ID, the ID is considered to continue, indicating a match between the input sensory measurement and the currently processed historical target. Values ​​outside the range are assigned a new ID, while IDs without sensory measurements within the range are eliminated. The number of historical matches N for the currently processed historical target is recorded. ID ;

[0030] Record the ID of the target perception measurement value that has been matched with the historical target being processed, by D match =D match '-(α*N ID D represents the Mahalanobis distance between the historical target being processed at the current moment and the target perception measurement. match 'D' is the Mahalanobis distance between the currently processed historical target and the perceptual measurement matched in the previous frame. match The value is the corrected Mahalanobis distance, and α is the correction factor coefficient.

[0031] In some optional implementations, the upper limit of the number of perceived measurements involved in the calculation for each historical target, and the construction of the Mahalanobis distance confirmation matrix, includes:

[0032] In the Mahalanobis distance matrix, if the number of perceived measurements within the coarse threshold range is greater than a preset number, the Mahalanobis distance values ​​between the historical target and each perceived measurement within the coarse threshold range are sorted, the largest Mahalanobis distance value is removed, and the ratio of the sum of all Mahalanobis distance values ​​after removing the largest Mahalanobis distance value to the sum of all Mahalanobis distance values ​​before removing the largest Mahalanobis distance value is calculated. If the ratio is less than the set threshold, all perceived measurements within the coarse threshold range are retained; otherwise, the second largest Mahalanobis distance value is removed, and the filtering and calculation continue. Values ​​in the Mahalanobis distance matrix at this time are set to 1 to obtain the Mahalanobis distance confirmation matrix.

[0033] In some optional implementations, dynamically adjusting the threshold parameters in the filtering algorithm based on the current data association method of the target within the sensing subdomain and the sensor data status includes:

[0034] During the Kalman filter tracking process, the estimated covariance P of the historical track is obtained through the correspondence between historical targets and measurements or the joint probability matrix, and the optimal state estimate of the historical target is updated by estimating the covariance P and the gain matrix K of the Kalman filter.

[0035] If the current historical target is changed from single-sensor tracking to multi-sensor tracking, that is, the source of the perception measurement value associated with the historical target becomes multiple sensors, or if the fluctuation of a certain dimension of the perception measurement value is greater than a preset threshold, or if it is known that the sensor detection has a weak area, then the variance of the perception measurement value is incorporated into the dynamic linear equation, and the variance of the perception measurement value is gradually corrected through the established attenuation interval, so that the variance of the perception measurement value changes according to different scenarios.

[0036] According to another aspect of the present invention, a multi-target fusion tracking device based on a sensing region is provided, comprising:

[0037] The perception domain segmentation module is used to divide the area to be perceived into multiple perception subdomains based on the lane line shape collected by the sensor, and each perception subdomain is classified according to the processing order.

[0038] The data association determination module is used to determine the environmental complexity in each sensing subdomain and to determine the data association method to be used in each sensing subdomain based on the environmental complexity.

[0039] The threshold adjustment module is used to dynamically adjust the threshold parameters in the filtering algorithm based on the current data association method of the target in the sensing subdomain and the sensor data status.

[0040] The fusion tracking module is used to perform data fusion tracking of target objects based on the threshold parameters corresponding to each sensing subdomain, and output the target state calculated by fusion.

[0041] According to another aspect of the present invention, a controller for a multi-target fusion tracking device based on a sensing region is provided.

[0042] According to another aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the methods described above.

[0043] In summary, compared with the prior art, the above-described technical solutions conceived by this invention can achieve the following beneficial effects:

[0044] Employing a composite fusion mechanism, different data association and fusion methods can be used to address various operating conditions, resulting in better robustness and higher overall algorithm efficiency. For complex scenarios, an improved joint probabilistic data association algorithm is introduced, and Mahalanobis distance is used to eliminate cases with low probabilities, improving operational efficiency and reducing computational load. For simple scenarios, a fast search method is introduced to quickly complete the data association function. In the subsequent tracking process, a dynamic parameter adjustment concept is introduced to respond in real time to the operating conditions during sensor perception, improving tracking stability. Attached Figure Description

[0045] Figure 1 This is a flowchart illustrating a multi-target fusion tracking method based on a sensing region provided in an embodiment of the present invention;

[0046] Figure 2 This is a schematic diagram of a lane shape provided by an embodiment of the present invention, wherein (a) is a straight lane, (b) is a diagonal lane, (c) shows the division of straight lanes, and (d) shows the division of diagonal lanes;

[0047] Figure 3 This is a flowchart illustrating a simple working condition for determining the data association method used in each sensing subdomain, as provided in an embodiment of the present invention.

[0048] Figure 4 It is a perceptual measurement ID management method;

[0049] Figure 5 This is a schematic diagram of threshold parameter adjustment provided in an embodiment of the present invention;

[0050] Figure 6 This is a schematic diagram of a multi-target fusion tracking device based on a sensing region provided in an embodiment of the present invention. Detailed Implementation

[0051] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention. Furthermore, the technical features involved in the various embodiments of this invention described below can be combined with each other as long as they do not conflict with each other.

[0052] In the embodiments of the present invention, "first," "second," etc., are used to distinguish different objects, rather than to describe a specific order or sequence.

[0053] In this invention, the perceived measurement value refers to the numerical value of various state characteristics returned by a sensor, such as a radar, after detecting a target.

[0054] To implement the technical solution of this invention, the required hardware equipment includes:

[0055] Camera sensor: used to perceive the state information of surrounding obstacles in the traffic environment, including their motion state, type and size; secondly, used to identify lane line information on the current road surface, including lane shape and type; can also serve as a data input for perceiving road conditions.

[0056] Millimeter-wave radar sensors are used to sense the motion information of surrounding dynamic obstacles in the current traffic environment, including dimensions such as position, speed and acceleration, as well as size information; they can be used as a data input for obstacle perception.

[0057] Vehicle signal sensors: Vehicle signal sensors mainly include vehicle speed sensors, yaw rate sensors and steering wheel angle sensors, which are used to sense the vehicle's motion status and can serve as a data input for vehicle motion status information.

[0058] Controller: Used to analyze the collected data, provide an algorithm running environment, and send control commands.

[0059] like Figure 1 The diagram shown is a flowchart illustrating a multi-target fusion tracking method based on a sensing region, according to an embodiment of the present invention. Figure 1 The method shown includes the following steps:

[0060] S1: Based on the lane line shape collected by the sensor, the area to be sensed is divided into multiple sensing subdomains, and each sensing subdomain is classified according to the processing order;

[0061] S2: Determine the complexity of the environment in each sensing subdomain, and determine the data association method to be used in each sensing subdomain based on the complexity of the environment;

[0062] S3: Dynamically adjust the threshold parameters in the filtering algorithm based on the current data association method of the target within the sensing subdomain and the sensor data status;

[0063] S4: Perform data fusion tracking of the target object based on the threshold parameters corresponding to each sensing subdomain, and output the target state obtained after fusion calculation.

[0064] In this embodiment of the invention, step S1 can be implemented in the following way:

[0065] S1.1: At the current moment, collect lane line data and construct a perception area in front of the vehicle based on the lane shape characteristics, such as... Figure 2 As shown, the width of the sensing area is equivalent to three lanes laterally, starting from the front axle of the vehicle and extending forward along the lane lines by a first preset distance (such as 120m in this embodiment of the invention). Figure 2 The image in (a) shows a straight lane. Figure 2 (b) shows a diagonal lane;

[0066] S1.2: The entire sensing area is divided into six sensing sub-domains: A, B, C1, C2, D1, and D2, based on the shape of the lane lines and the longitudinal distance. Figure 2 Image (c) shows the division of straight lanes. Figure 2The diagram (d) shows the division of the diagonal lanes. Each sensing subdomain is the width of one lane and extends along a second preset distance (e.g., 60m in this embodiment). The letter order represents the importance of the current sensing subdomain. For example, under normal vehicle driving conditions, the area within 60m of the current lane involves the risk of longitudinal collision and is therefore designated as Zone A. The area from 60 to 120m is a potential longitudinal collision zone and is designated as Zone B. The left and right sides are designated as C1, C2 and D1, D2 respectively based on the longitudinal distance.

[0067] S1.3: Fill the queue (Queue) with the set priority order of the perception subdomains. The processing perception subdomain queue is as follows:

[0068] Queue area ={D2 area D1 area C2 area C1 area B area A area}

[0069] Following the queue entry and exit criteria, the subdomains at the tail of the queue are processed first, and then popped out in sequence until the queue is empty.

[0070] In this embodiment of the invention, step S2 can be implemented in the following way:

[0071] S2.1: For the current perceptual subdomain to be processed, first calculate the complexity of the current scene within the perceptual subdomain, therefore, introduce the environmental complexity influence factor σ:

[0072]

[0073] Where n is the number of historical targets in the current scene within the perception subdomain. The more targets there are, the higher the complexity of the current perception subdomain. A complexity threshold T is set to be used. constant The judgment condition is:

[0074]

[0075] The result is used to determine the complexity of the scene within the current perception subdomain.

[0076] S2.2: As Figure 3As shown, when the result is 0, the current sensing subdomain is in a simple working condition. The strategy adopted is as follows: First, based on the range of the sensing subdomain, capture the sensing measurement data of the corresponding sensors (such as cameras and millimeter-wave radar) within that range, and construct a sensing measurement list of the sensors within the sensing subdomain; second, find the historical targets that already exist within that range, and construct an ordered historical target list based on the length of time they have existed; then, perform the matching step: 1. Calculate the Mahalanobis distance D between each historical target and each sensing measurement value. match 2. For historical targets, introduce the historical matching count N. ID This is used to correct distance values. The method involves establishing an ID management mechanism for the input sensory measurements. An ID is assigned to the initial input measurement. The range of distance and velocity variation is defined according to the equation of uniform motion. If the input measurement in the next frame falls within this range, the ID is considered to continue. If it exceeds the range, a new ID is generated. IDs for measurements without measurement within the range are deleted. Figure 4 As shown, the system then records the perceptual measurement IDs that have been matched with the current target in the past, and introduces a correction factor coefficient α and the number of matches N. ID The Mahalanobis distance between the current historical target and the matched perceptual measurement is:

[0077] D match =D match '-(α*N ID )

[0078] D match 'D' is the Mahalanobis distance between the currently processed historical target and the perceptual measurement matched in the previous frame. match These are the corrected Mahalanobis distance values.

[0079] 3. Select the measurement corresponding to the minimum distance value from the perception measurement list in the order of the historical target list;

[0080] 4. Establish the correspondence between historical targets and measurements.

[0081] S2.3: When result is 1, the current sensing subdomain is in a complex operating condition.

[0082] 1. As in step S2.2, acquire sensor measurement data from sensors within the sensing subdomain and construct a sensing measurement list. Identify historical targets that already exist within the sensing subdomain and construct a historical target list based on the duration of each historical target's existence. For each historical target, set a coarse threshold covering all dimensions (coarse means the threshold is set relatively large, only to filter sensing measurement values ​​that cannot be associated) to avoid invalid Mahalanobis distance calculations. Calculate the Mahalanobis distance between the historical target and the sensing measurement values ​​in the sensing measurement list within the sensing subdomain that meet their respective coarse threshold ranges, and construct a Mahalanobis distance matrix. Set the position of measurements outside the coarse threshold range in the Mahalanobis distance matrix to 0.

[0083] 2. To reduce the overall algorithm complexity, the upper limit of the number of perceived measurements involved in the calculation for each historical target is controlled by setting a threshold T a priori. dist For historical targets, if the number of perceived measurements within the coarse threshold range is greater than 3, sort the Mahalanobis distance values ​​between the historical targets and the perceived measurements, remove the maximum value, and calculate:

[0084]

[0085] Dist i Here, α represents the Mahalanobis distance value, and m is the number of Mahalanobis distance values. If α is less than T... dist If the value is within the range, all perceived measurement values ​​are retained; otherwise, the second largest Mahalanobis distance value is removed, and α is calculated by filtering according to the above formula.

[0086] 3. Set all values ​​in the Mahalanobis distance matrix to 1 to obtain the Mahalanobis distance confirmation matrix;

[0087] 4. Generate feasible event clusters based on the confirmation matrix: For the Mahalanobis distance confirmation matrix, first confirm the position of all 1 elements, and then, according to the basic rules of feasible events (one perceptual measurement value corresponds to one historical target), generate a matrix of all feasible events (feasible events are presented in matrix form, with only one non-zero value in each row and column, representing a possible matching result). For example, if the confirmation matrix is ​​as follows:

[0088]

[0089] Obtain the feasible event matrix:

[0090] and By analogy, feasible event clusters can be obtained;

[0091] 5. Calculate the correlation probability of each event, statistically calculate the joint probability of each historical target, and obtain the joint probability matrix.

[0092] In this embodiment of the invention, step S3 can be implemented in the following way:

[0093] S3.1: In the subsequent Kalman filter tracking process, since obtaining the historical target-measurement correspondence or joint probability matrix in S2 is for obtaining the estimated covariance P of the historical trajectory, and the Kalman filter gain matrix K is used to update the optimal state estimate of the historical target: (P t (Predicted from the previous frame P)

[0094]

[0095] H is a parameter of the measurement system, and the estimated variance R of the sensed measurement should also be adjusted. The following three cases should be identified as needed:

[0096] 1. When the target is tracked by a single sensor and then by multiple sensors, that is, when the source of the perception measurement value associated with the historical target becomes multiple sensors, the tracking events are different due to the observation principles of multiple sensors, resulting in a large error at the beginning.

[0097] 2. When a certain dimension of the perceived measurement value fluctuates significantly, this fluctuation will affect the fusion result;

[0098] 3. In areas where some sensors have weak detection capabilities, such as at the edge of radar's energy range.

[0099] S3.2: Detecting the situation described in S3.1, a dynamic adjustment mechanism is introduced to cope with different working scenarios, incorporating the variance of the sensed measurement values ​​into the dynamic linear equation:

[0100] R Dynamic =f(X,R) min ,R normal, R max )

[0101] X represents distance, R min R represents the minimum variance of the perceived measurement. normal R represents the standard value of the variance of the perceived measurement. max This represents the maximum value of the variance of the perceived measurement.

[0102] like Figure 5 As shown, in scenarios where parameters change abruptly, the perceived measurement variance R enters different transformation models according to the settings. By gradually correcting the parameter value of R through the established attenuation interval, the optimal estimation of the system becomes more stable and accurate, reducing the disturbances caused by abnormal changes.

[0103] In this embodiment of the invention, the final results obtained from the fusion in S4 can be assembled into a target list, which can then be pushed to subsequent modules and prepared for data processing in the next frame.

[0104] The technical solution of this invention includes: confirming the division and formation of the sensing region and sensing sub-domain queues, and setting data association and fusion strategies for each sub-domain. In simple scenarios, a fast data association method is designed based on historical target and measurement information. In complex scenarios, an improved joint probability data association algorithm is designed, reducing the amount of data computation and improving the stability of fusion. For subsequent tracking, a dynamic parameter adjustment method is designed, and a linear influence formula that changes with the attenuation region is set to address different matching situations and sensor sensing states during the tracking process.

[0105] like Figure 6 The diagram shown is a schematic diagram of a multi-target fusion tracking device based on a sensing region provided in an embodiment of the present invention, including:

[0106] The perception domain segmentation module 601 is used to divide the area to be perceived into multiple perception subdomains according to the shape of the lane lines collected by the sensor, and each perception subdomain is classified according to the processing order.

[0107] The data association determination module 602 is used to determine the environmental complexity in each sensing subdomain and determine the data association method to be used in each sensing subdomain based on the environmental complexity.

[0108] The threshold adjustment module 603 is used to dynamically adjust the threshold parameters in the filtering algorithm according to the current data association method of the target in the sensing subdomain and the sensor data status.

[0109] The fusion tracking module 604 is used to perform data fusion tracking of the target object based on the threshold parameters corresponding to each sensing subdomain, and output the target state calculated by fusion.

[0110] The specific implementation methods of each module can be referred to the description of the above method embodiments, and the embodiments of the present invention will not be repeated.

[0111] It should be noted that, depending on the implementation needs, the various steps / components described in this application can be broken down into more steps / components, or two or more steps / components or parts of the operation of steps / components can be combined into new steps / components to achieve the purpose of this invention.

[0112] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-target fusion tracking method based on a perception region, characterized in that, include: Based on the lane line shape collected by the sensor, the area to be sensed is divided into multiple sensing subdomains, and each sensing subdomain is classified according to the processing order. The complexity of the environment is determined in each sensing subdomain, and the data association method used in each sensing subdomain is determined based on the complexity of the environment; The threshold parameters in the filtering algorithm are dynamically adjusted based on the current data association method of the target within the sensing subdomain and the sensor data status. Data fusion tracking of the target object is performed based on the threshold parameters corresponding to each sensing subdomain, and the target state calculated by fusion is output. The step of determining the environmental complexity in each sensing subdomain and using the environmental complexity to determine the data association method used in each sensing subdomain includes: For the current perceptual subdomain to be processed in the queue, calculate the complexity of the current scene in the perceptual subdomain. The more objects there are in the current scene of the perceptual subdomain, the higher the complexity of the perceptual subdomain. When the sensing subdomain is determined to be a simple working condition, the sensing measurement data of the sensors within the sensing subdomain is obtained to construct a sensing measurement list. Historical targets that already exist within the sensing subdomain are identified. A historical target list is constructed based on the length of time each historical target has existed. Each historical target in the historical target list is matched with each sensing measurement value in the sensing measurement list to construct the correspondence between each historical target and the sensing measurement value. When a complex working condition is determined within the sensing subdomain, the sensing measurement data of the sensors within the sensing subdomain are acquired to construct a sensing measurement list. Historical targets that already exist within the sensing subdomain are identified, and a historical target list is constructed based on the duration of existence of each historical target. For each historical target, a coarse threshold covering various dimensions is set. The Mahalanobis distance between each historical target and the sensing measurement values ​​in the sensing measurement list within the sensing subdomain that meet the respective coarse threshold range is calculated, and a Mahalanobis distance matrix is ​​constructed. Positions in the Mahalanobis distance matrix that are not within the coarse threshold range are set to 0. The upper limit of the number of sensing measurement values ​​that each historical target participates in the calculation is controlled, and a Mahalanobis distance confirmation matrix is ​​constructed. For the Mahalanobis distance confirmation matrix, all positions that are target values ​​are confirmed. Based on the rule that one perception measurement value corresponds to one historical target, a matrix of all feasible events is generated to obtain a cluster of feasible events. The feasible events are presented in the form of a matrix, and each row and column can only contain one target value. Calculate the association probability of each event in the feasible event cluster, statistically calculate the joint probability of each historical target, and obtain the joint probability matrix.

2. The method according to claim 1, characterized in that, The process of dividing the area to be sensed into multiple sensing sub-domains based on the lane line shape collected by the sensors includes: At the current moment, lane line data is collected, and a perception area is constructed in front of the vehicle based on the shape of the lane line. The width of the perception area is the total lane width, and the longitudinal direction starts from the front axle of the vehicle and extends forward along the lane line by a first preset distance. The sensing area is divided into multiple sensing sub-domains based on the shape of the lane lines and the longitudinal distance. The width of each sensing sub-domain is the width of one lane, and the longitudinal distance is along the lane area with a second preset length. The processing order of the set sensing subdomains is filled into the queue to obtain the processing subdomain queue of the sensing region.

3. The method according to claim 2, characterized in that, Depend on Computational perception subdomain complexity within the current scene , , n To perceive the number of target objects within the current scene of the subdomain, For complex thresholds, =0 indicates that the sensing subdomain is in a simple working condition. =1 indicates that the sensing subdomain is in a complex working condition.

4. The method according to claim 2 or 3, characterized in that, The step of matching each historical target in the historical target list with each perception measurement value in the perception measurement list to construct the correspondence between each historical target and perception measurement value includes: Calculate the Mahalanobis distance between each historical target and each perceived measurement; Based on the target perception measurement value that matches the previous frame of the historical target being processed, the Mahalanobis distance value between the historical target being processed at the current moment and the target perception measurement value is corrected by the number of historical matches. Following the order of historical targets in the historical target list, the perceptual measurement values ​​corresponding to the minimum Mahalanobis distance between the historical targets are selected sequentially from the perceptual measurement list to construct the correspondence between the historical targets and the perceptual measurement values ​​corresponding to the minimum Mahalanobis distance.

5. The method according to claim 4, characterized in that, The step of correcting the Mahalanobis distance value between the current historical target and the target perception measurement value based on the target perception measurement value matched with the previous frame of the currently processed historical target, by adjusting the historical matching count, includes: Each initial input sensory measurement is assigned an ID, and the distance and velocity variation ranges for each ID are defined. If an input sensory measurement in the next frame falls within the range of a given ID, the ID is considered to continue, indicating a match between the input sensory measurement and the currently processed historical target. Values ​​outside the range are assigned a new ID, while IDs without sensory measurements within the range are eliminated. The number of historical matches N for the currently processed historical target is recorded. ID ; Record the ID of the target perception measurement value that has been matched with the historical target being processed. The Mahalanobis distance between the historical target and the target perception measurement value processed at the current moment. The Mahalanobis distance is the distance between the currently processed historical target and the perceptual measurement value matched in the previous frame. The value is the corrected Mahalanobis distance, and α is the correction factor coefficient.

6. The method according to claim 2 or 3, characterized in that, The upper limit of the number of perceived measurements used in the calculation for each historical target is controlled, and a Mahalanobis distance confirmation matrix is ​​constructed, including: In the Mahalanobis distance matrix, if the number of perceived measurements within the coarse threshold range is greater than a preset number, the Mahalanobis distance values ​​between the historical target and each perceived measurement within the coarse threshold range are sorted, the largest Mahalanobis distance value is removed, and the ratio of the sum of all Mahalanobis distance values ​​after removing the largest Mahalanobis distance value to the sum of all Mahalanobis distance values ​​before removing the largest Mahalanobis distance value is calculated. If the ratio is less than the set threshold, all perceived measurements within the coarse threshold range are retained; otherwise, the second largest Mahalanobis distance value is removed, and the filtering and calculation continue. Values ​​in the Mahalanobis distance matrix at this time are set to 1 to obtain the Mahalanobis distance confirmation matrix.

7. The method according to claim 6, characterized in that, The step of dynamically adjusting the threshold parameters in the filtering algorithm based on the current data association method of the target within the sensing subdomain and the sensor data status includes: During the Kalman filter tracking process, the estimated covariance P of the historical track is obtained through the correspondence between historical targets and measurements or the joint probability matrix, and the optimal state estimate of the historical target is updated by estimating the covariance P and the gain matrix K of the Kalman filter. If the current historical target is changed from single-sensor tracking to multi-sensor tracking, that is, the source of the perception measurement value associated with the historical target becomes multiple sensors, or if the fluctuation of a certain dimension of the perception measurement value is greater than a preset threshold, or if it is known that the sensor detection has a weak area, then the variance of the perception measurement value is incorporated into the dynamic linear equation, and the variance of the perception measurement value is gradually corrected through the established attenuation interval, so that the variance of the perception measurement value changes according to different scenarios.

8. A multi-target fusion tracking device based on a sensing region, characterized in that, include: The perception domain segmentation module is used to divide the area to be perceived into multiple perception subdomains based on the lane line shape collected by the sensor, and each perception subdomain is classified according to the processing order. The data association determination module is used to determine the environmental complexity in each sensing subdomain and to determine the data association method to be used in each sensing subdomain based on the environmental complexity. The threshold adjustment module is used to dynamically adjust the threshold parameters in the filtering algorithm based on the current data association method of the target in the sensing subdomain and the sensor data status. The fusion tracking module is used to perform data fusion tracking of target objects based on the threshold parameters corresponding to each sensing subdomain, and output the target state obtained after fusion calculation; The data association determination module is specifically used to perform the following operations: For the current perceptual subdomain to be processed in the queue, calculate the complexity of the current scene in the perceptual subdomain. The more objects there are in the current scene of the perceptual subdomain, the higher the complexity of the perceptual subdomain. When the sensing subdomain is determined to be a simple working condition, the sensing measurement data of the sensors within the sensing subdomain is obtained to construct a sensing measurement list. Historical targets that already exist within the sensing subdomain are identified. A historical target list is constructed based on the length of time each historical target has existed. Each historical target in the historical target list is matched with each sensing measurement value in the sensing measurement list to construct the correspondence between each historical target and the sensing measurement value. When a complex working condition is determined within the sensing subdomain, the sensing measurement data of the sensors within the sensing subdomain are acquired to construct a sensing measurement list. Historical targets that already exist within the sensing subdomain are identified, and a historical target list is constructed based on the duration of existence of each historical target. For each historical target, a coarse threshold covering various dimensions is set. The Mahalanobis distance between each historical target and the sensing measurement values ​​in the sensing measurement list within the sensing subdomain that meet the respective coarse threshold range is calculated, and a Mahalanobis distance matrix is ​​constructed. Positions in the Mahalanobis distance matrix that are not within the coarse threshold range are set to 0. The upper limit of the number of sensing measurement values ​​that each historical target participates in the calculation is controlled, and a Mahalanobis distance confirmation matrix is ​​constructed. For the Mahalanobis distance confirmation matrix, all positions that are target values ​​are confirmed. Based on the rule that one perception measurement value corresponds to one historical target, a matrix of all feasible events is generated to obtain a cluster of feasible events. The feasible events are presented in the form of a matrix, and each row and column can only contain one target value. Calculate the association probability of each event in the feasible event cluster, statistically calculate the joint probability of each historical target, and obtain the joint probability matrix.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images