An abnormal vehicle detection and evidence chain construction method based on trajectory analysis

Abstract

The application relates to the technical field of intelligent traffic data processing, and discloses an abnormal vehicle detection and evidence chain construction method based on trajectory analysis. By establishing and normalizing a vehicle trajectory object, the time interval, ground distance and passing speed of adjacent trajectory records are calculated, trajectory segments are divided, and abnormal intensity and abnormal types are obtained based on the road compression amount, low-speed retention amount and time cavity amount; candidate evidence nodes are generated by screening and activating segments, candidate evidence chains are constructed and scored in combination with time, space, abnormal types and identity credibility, the optimal abnormal evidence chain is selected to determine the abnormal vehicle and output the result, so that the problems of single-point abnormality, single-section speed abnormality or fixed threshold judgment prone to false positives, false negatives and insufficient explanation are solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation data processing technology, specifically to a method for abnormal vehicle detection and evidence chain construction based on trajectory analysis. Background Technology

[0002] With the development of intelligent transportation systems, road video sensing equipment, vehicle positioning terminals, and traffic operation monitoring platforms, vehicle latitude and longitude trajectory data has become an important data source for abnormal vehicle identification. In existing technologies, abnormal vehicle detection is typically based on data such as vehicle location, speed changes, dwell time, route deviation, and road checkpoint passage records. For example, some methods identify abnormal acceleration, abnormal low speed, or abnormal stopping behavior by setting a preset speed threshold; some methods determine whether a vehicle is abnormal by comparing the deviation between the vehicle's actual trajectory and preset road trajectories or historical regular trajectories; and some methods utilize time differences, distance differences, or checkpoint passage sequences between consecutive positioning points to identify suspected detours, delays, or missing trajectories.

[0003] However, existing technologies still have several shortcomings in practical applications. First, existing methods mostly rely on manually set thresholds or fixed empirical rules. Trajectory data varies greatly under different road environments, sampling frequencies, and traffic conditions, and fixed thresholds are difficult to adapt to the unique trajectory characteristics of vehicles, easily leading to false alarms or missed alarms. Second, existing methods typically use a single trajectory point, a single time period, or a single anomaly indicator as the basis for judgment, making it difficult to simultaneously reflect multiple anomalies such as distance compression, low-speed stagnation, and time gaps in trajectory segments, resulting in a lack of stability and interpretability in anomaly identification results. Third, vehicle trajectory data may contain location jumps, uneven collection intervals, and fluctuations in identity recognition confidence. If existing technologies directly use maximum speed, single-point distance, or single time interval as the basis for judgment, they are easily affected by individual extreme data. In addition, most existing abnormal vehicle detection methods only output whether a vehicle is abnormal or an anomaly score, lacking a mechanism for constructing the correlation between abnormal trajectory segments. For vehicles with multiple anomaly segments scattered throughout, existing technologies struggle to determine whether the anomaly segments are continuous in time, reachable in space, and consistent in anomaly type, and it is also difficult to form an evidence chain that reflects the evolution of abnormal behavior.

[0004] Therefore, this case aims to propose an abnormal vehicle detection and evidence chain construction method based on trajectory analysis. First, the vehicle trajectory records are cleaned, sorted, standardized and fragmented. Then, abnormal behaviors such as route compression, low-speed stagnation and time gaps are extracted from each trajectory segment. Candidate evidence nodes are further screened, and multiple candidate evidence nodes are connected into an evidence chain based on temporal continuity, spatial accessibility, consistency of abnormal type and identity credibility. Summary of the Invention

[0005] This invention provides a method for abnormal vehicle detection and evidence chain construction based on trajectory analysis, which helps to solve the problems mentioned in the background art.

[0006] This invention provides the following technical solution: a method for abnormal vehicle detection and evidence chain construction based on trajectory analysis, comprising:

[0007] Vehicle trajectory objects within the detection area are established. Latitude and longitude trajectory records are organized according to collection time, retaining valid vehicle objects and standardizing latitude and longitude, collection time, and vehicle object identification confidence values. The time interval, surface distance, and travel speed between adjacent trajectory records of valid vehicle objects are calculated, and trajectory segments are divided according to the number of trajectory records. The distance compression, low-speed dwell time, and time hole of each trajectory segment are calculated to obtain anomaly intensity, and the anomaly type corresponding to the dominant anomaly is recorded. Trajectory segments in an active state are selected based on the mean anomaly intensity of segments within the vehicle and the average deviation of segment anomaly intensity, generating candidate evidence nodes. For vehicles containing candidate evidence nodes, calculations are performed... Calculate the time reference, distance reference, and reachable traffic speed between candidate evidence nodes; for candidate evidence node pairs of the same vehicle, calculate the combined results of time continuity, spatial reachability, anomaly type consistency, and identity credibility to obtain the splicing credibility between candidate evidence nodes; construct candidate evidence chains according to the numerical order of the center time of candidate evidence nodes and the splicing credibility between candidate evidence nodes, calculate the candidate evidence chain score, and select the optimal abnormal evidence chain; determine abnormal vehicles based on the optimal abnormal evidence chain score, the average strength of vehicle evidence chains within the chain-constructed vehicle set, and the average deviation of vehicle evidence chain strength within the chain-constructed vehicle set, and output the abnormal vehicle detection results.

[0008] Optionally, the step of establishing vehicle trajectory objects within the area to be detected, organizing latitude and longitude trajectory records according to the collection time, retaining valid vehicle objects, and standardizing latitude and longitude, collection time, and vehicle object identification confidence values ​​specifically includes:

[0009] An initial vehicle object set is established for all vehicles within the detection area, and the total number of initial vehicles is recorded. The trajectory records for each vehicle are arranged in ascending order of acquisition time. Trajectory records output from the upstream trajectory aggregation platform and associated with vehicle object numbers by the vehicle object recognition processing unit are received. For each trajectory record, the vehicle number, longitude, latitude, acquisition time, and vehicle object recognition confidence value are extracted. The vehicle object recognition confidence value is recorded as the degree of matching confidence between the trajectory record and the corresponding vehicle number, and the vehicle object recognition confidence value is limited to a range of zero to one. When multiple trajectory records exist at the same acquisition time, the vehicle object recognition confidence values ​​of each trajectory record are compared, and the trajectory record with the highest vehicle object recognition confidence value is retained. When multiple trajectory records with the same vehicle object recognition confidence value exist at the same acquisition time, the original import sequence number of each trajectory record is compared, and the trajectory record with the smallest original import sequence number is retained. The retained trajectory records are renumbered in strictly ascending order of acquisition time. The number of renumbered trajectory records for each vehicle is counted, and vehicles with at least two renumbered trajectory records are included in the valid vehicle objects. For each vehicle in the set of valid vehicle objects, a trajectory record set is created, with the trajectory records arranged in strictly ascending order of acquisition time. For each trajectory record, the longitude is converted to radians, the latitude is converted to radians, and the acquisition time is converted to seconds. The vehicle object identification confidence value for each trajectory record is also recorded. For any two latitude and longitude points, the square of the sine of half the difference in radians between the two latitudes is first calculated, followed by the cosine of the first latitude, the cosine of the second latitude, and the radians of the two longitudes. The product of the square of half the difference in latitude arc length is used to form the auxiliary value of the surface arc distance. The square root of the auxiliary value of the surface arc distance is then taken and the arcsine of the two latitude arc length differences is added to the product of the square of the first latitude cosine value, the second latitude cosine value, and the square of the half sine of the difference in longitude arc length. The result is then multiplied by two and the average radius of the Earth to form the surface arc distance between the two latitude and longitude points, where the average radius of the Earth is taken as 6,371,000 meters. The length reference constant is set to one meter, and the time reference constant is set to one second.

[0010] Optionally, the calculation of the time interval, surface distance, and traffic speed of adjacent trajectory records of valid vehicle objects, and the division of trajectory segments according to the number of trajectory records, specifically includes:

[0011] For each vehicle's adjacent trajectory records, the time interval between adjacent trajectory records is obtained by subtracting the acquisition time value of the previous trajectory record from the acquisition time value of the later trajectory record; for each vehicle's adjacent trajectory records, the surface distance between the previous and subsequent trajectory records is calculated according to the method of calculating the surface arc distance of latitude and longitude; for each vehicle's adjacent trajectory records, the surface distance between adjacent trajectory records is divided by the time interval between adjacent trajectory records to obtain the travel speed between adjacent trajectory records; the number of trajectory records for each vehicle is reduced by one to obtain the length of the first candidate segment; the preset minimum segment length is compared with the length of the first candidate segment, and the larger of the two is taken as the length of the second candidate segment; the first candidate... The segment length is compared with the second candidate segment length, and the smaller value is taken as the segment length of the current vehicle; the number of trajectory records is reduced by one and divided by the segment length, and the result is rounded up to obtain the total number of trajectory segments for the current vehicle; the starting trajectory record number of each trajectory segment is calculated according to the trajectory segment number and the segment length; the candidate ending trajectory record number of each trajectory segment is calculated according to the trajectory segment number and the segment length, and the candidate ending trajectory record number is compared with the number of trajectory records for the current vehicle, and the smaller value is taken as the ending trajectory record number of the current trajectory segment; the trajectory record is extracted according to the starting trajectory record number and the ending trajectory record number of each trajectory segment to form the trajectory segment of the corresponding vehicle.

[0012] Optionally, the calculation of the path compression, low-speed retention, and time hole of each trajectory segment to obtain the anomaly intensity, and the recording of the anomaly type corresponding to the dominant anomaly quantity, specifically includes:

[0013] For each trajectory segment of each vehicle, the surface distance between adjacent trajectory records within the current trajectory segment is accumulated to obtain the cumulative distance of the segment; for each trajectory segment of each vehicle, the surface arc distance between the endpoints of the starting and ending trajectory records of the current trajectory segment is calculated according to the calculation method of latitude and longitude surface arc distance; the endpoint surface arc distance is subtracted from the cumulative distance of the segment, and the difference between the cumulative distance of the segment and the endpoint surface arc distance is compared with zero, and the larger value is taken as the distance difference; the cumulative distance of the segment is added to the length reference constant to obtain the normalized denominator of the distance; the distance difference is divided by the normalized denominator of the distance to obtain the distance compression of the current trajectory segment; the same vehicle... All adjacent traffic speeds of vehicles are sorted in ascending order of numerical value, and the lower median traffic speed is selected. Adjacent traffic speeds within each trajectory segment are compared one by one. When an adjacent traffic speed is less than or equal to the lower median traffic speed, a low-speed count is recorded; when an adjacent traffic speed is greater than the lower median traffic speed, no low-speed count is added. The low-speed count is divided by the number of adjacent movements compared within the current trajectory segment to obtain the low-speed dwell time of the current trajectory segment. The time intervals between all adjacent trajectory records for the same vehicle are summed, and the sum is divided by the number of adjacent trajectory record time intervals to obtain the average of the adjacent trajectory record time intervals. The time intervals between adjacent trajectory records within each trajectory segment are... Each time interval is processed individually. The average time interval between adjacent trajectory records is subtracted from the average time interval between adjacent trajectory records. The difference between the two is compared with zero, and the larger value is taken as the time interval excess. The time interval between adjacent trajectory records is added to the time base constant to obtain the time normalized denominator. The time interval excess is divided by the time normalized denominator to obtain the time hole sub-quantity of a single adjacent trajectory record time interval. All time hole sub-quantities within the same trajectory segment are summed, and the sum is divided by the number of adjacent motion quantities involved in the calculation within the current trajectory segment to obtain the time hole quantity of the current trajectory segment. The path compression, low-speed stagnation, and time void of the same trajectory segment are added together, and the sum is divided by three to obtain the anomaly intensity of the current trajectory segment. When the path compression is greater than or equal to the low-speed stagnation and the path compression is greater than or equal to the time void, the anomaly type of the current trajectory segment is recorded as path compression-dominated anomaly. When the path compression is less than the low-speed stagnation and the low-speed stagnation is greater than or equal to the time void, the anomaly type of the current trajectory segment is recorded as low-speed stagnation-dominated anomaly. When the path compression is less than the low-speed stagnation and the low-speed stagnation is less than the time void, the anomaly type of the current trajectory segment is recorded as time void-dominated anomaly.

[0014] Optionally, the step of filtering active trajectory segments based on the mean of segment anomaly intensity and the mean deviation of segment anomaly intensity to generate candidate evidence nodes specifically includes:

[0015] For each vehicle, sum the anomaly intensities of all segments and divide the sum by the total number of trajectory segments for that vehicle to obtain the mean anomaly intensity of segments within the vehicle. For each trajectory segment, subtract the mean anomaly intensity of segments within the vehicle from the anomaly intensity of the current trajectory segment, and take the absolute value of the difference between the anomaly intensity of the current trajectory segment and the mean anomaly intensity of segments within the vehicle to obtain the segment anomaly intensity deviation. Sum the anomaly intensity deviations of all segments for the same vehicle and divide the sum by the total number of trajectory segments for that vehicle to obtain the average anomaly intensity deviation of segments. Add the mean anomaly intensity of segments within the vehicle and the average anomaly intensity deviation of segments to form a candidate activation threshold. For each trajectory segment, compare the anomaly intensity with the candidate... The system compares activation thresholds. When the anomaly intensity of a trajectory segment is greater than zero and greater than or equal to the candidate activation threshold, the candidate activation marker for the current trajectory segment is recorded as active. When the anomaly intensity of a trajectory segment is equal to zero, the candidate activation marker for the current trajectory segment is recorded as inactive. When the anomaly intensity of a trajectory segment is greater than zero and less than the candidate activation threshold, the candidate activation marker for the current trajectory segment is recorded as inactive. For each trajectory segment, the acquisition time value of the starting trajectory record and the acquisition time value of the ending trajectory record are added together, and the result is divided by two to obtain the center time value of the current trajectory segment. For trajectory segments in the active state, all trajectories within the current trajectory segment are... The longitude records are summed, and the sum is divided by the number of trajectory records in the current trajectory segment to obtain the center longitude. For an active trajectory segment, the latitude records of all trajectory records in the current trajectory segment are summed, and the sum is divided by the number of trajectory records in the current trajectory segment to obtain the center latitude. For an active trajectory segment, the vehicle object identification confidence values ​​of all trajectory records in the current trajectory segment are summed, and the sum is divided by the number of trajectory records in the current trajectory segment to obtain the identity credibility. All active trajectory segments are sorted in ascending order of their center time values. When two active trajectory segments have the same center time value, they are sorted in ascending order of the starting trajectory record number of the active segment. For each active trajectory segment, candidate evidence nodes are generated. Each candidate evidence node includes its time interval, center time value, center longitude, center latitude, anomaly type, anomaly intensity, and identity credibility. The time interval of the candidate evidence node consists of the acquisition time values ​​of the starting trajectory record and the ending trajectory record of the current trajectory segment. The candidate evidence node sequence number and the original trajectory segment sequence number corresponding to each candidate evidence node are recorded according to the arrangement result. The number of candidate evidence nodes for each vehicle is recorded. When there are no active trajectory segments for the same vehicle, the number of candidate evidence nodes for the current vehicle is recorded as zero, and no candidate evidence node is generated.

[0016] Optionally, for vehicles containing candidate evidence nodes, calculating the time reference between candidate evidence nodes, the distance reference between candidate evidence nodes, and the achievable passage speed specifically includes:

[0017] Vehicles with at least one candidate evidence node are included in the chain-constructed vehicle set; the number of elements in the chain-constructed vehicle set is compared with one, and the larger value is used as the effective denominator for the chain-constructed vehicle set; when the chain-constructed vehicle set is empty, the summation result on the chain-constructed vehicle set is recorded as zero; for each vehicle in the chain-constructed vehicle set, all candidate evidence nodes are sorted in ascending order according to their center time values ​​to form a set of candidate evidence nodes for the same vehicle; for each vehicle in the chain-constructed vehicle set, the current vehicle's lower median traffic speed is subtracted from each adjacent traffic speed, and the absolute value of the difference between each adjacent traffic speed and the current vehicle's lower median traffic speed is taken to obtain the traffic speed deviation; the traffic speed deviations of the same vehicle are summed, and the summation result is divided by the number of adjacent traffic speeds of the current vehicle to obtain the average traffic speed deviation; the current vehicle's lower median traffic speed is added to the average traffic speed deviation to form the achievable traffic speed screening upper limit; adjacent traffic speeds not greater than the achievable traffic speed screening upper limit are selected from all adjacent traffic speeds of the current vehicle, and... The maximum value in the screening results is taken as the achievable passage speed of the current vehicle. For vehicles with more than one candidate evidence node in the chain-constructed vehicle set, the difference in the center time values ​​of adjacent candidate evidence nodes is calculated one by one, and all center time value differences are summed. The sum of the center time value differences of adjacent candidate evidence nodes is divided by the number of candidate evidence nodes of the current vehicle to obtain the time reference between the candidate evidence nodes of the current vehicle. When the number of candidate evidence nodes of the same vehicle is one, the sum of the center time value differences of adjacent candidate evidence nodes is recorded as zero. For vehicles with more than one candidate evidence node in the chain-constructed vehicle set, the center position arc distance of adjacent candidate evidence nodes is calculated one by one according to the calculation method of latitude and longitude surface arc distance, and all center position arc distances are summed. The sum of the center position arc distances of adjacent candidate evidence nodes is divided by the number of candidate evidence nodes of the current vehicle to obtain the distance reference between the candidate evidence nodes of the current vehicle. When the number of candidate evidence nodes of the same vehicle is one, the sum of the center position arc distances of adjacent candidate evidence nodes is recorded as zero.

[0018] Optionally, the calculation of the combined results of temporal continuity, spatial reachability, anomaly type consistency, and identity credibility for candidate evidence node pairs of the same vehicle yields the concatenation credibility between candidate evidence nodes, specifically including:

[0019] For each vehicle in the chain-constructed vehicle set, select preceding and subsequent candidate evidence nodes in order of their arrangement numbers, ensuring that the arrangement number of the preceding candidate evidence node is less than that of the subsequent candidate evidence node. Subtract the center time value of the preceding candidate evidence node from the center time value of the subsequent candidate evidence node to obtain the difference in center time values ​​between them. Divide this difference in center time values ​​by the sum of the time base and the time base constant between the candidate evidence nodes of the current vehicle to form the result. The time continuity quotient is calculated by taking the negative of the time continuity quotient and then performing a natural exponential operation to obtain the time continuity between preceding and subsequent candidate evidence nodes. Following the method for calculating latitude and longitude surface arc distances, the center position arc distance between the center longitude and center latitude of the preceding and subsequent candidate evidence nodes is calculated. The current achievable traffic speed of the vehicle is multiplied by the difference in center time values ​​between the preceding and subsequent candidate evidence nodes to form the reachable distance between candidate evidence nodes. The reachable distance between candidate evidence nodes is subtracted from the center position arc distance, and the center position arc distance is then calculated. The difference between the position arc distance and the reachable distance between candidate evidence nodes is compared with zero, and the larger value is taken as the excess reachable distance. The excess reachable distance is divided by the sum of the distance benchmark and length benchmark constant between candidate evidence nodes of the current vehicle to form the spatial reachability quotient. The spatial reachability quotient is then inversely calculated and subjected to natural exponentiation to obtain the spatial reachability between the preceding and subsequent candidate evidence nodes. The anomaly types of the preceding and subsequent candidate evidence nodes are compared. When they are the same, the anomaly type consistency is recorded as one. The anomaly types of the preceding and subsequent candidate evidence nodes are compared. When the anomaly type of the normal type differs from that of the subsequent candidate evidence node, the anomaly type of the preceding candidate evidence node is subtracted from that of the subsequent candidate evidence node, and the absolute value of the difference is taken. The absolute value of the difference is added to one, and the reciprocal of the sum is taken to obtain the anomaly type consistency between the preceding and subsequent candidate evidence nodes. The temporal continuity, spatial reachability, anomaly type consistency, identity credibility of the preceding candidate evidence node, and identity credibility of the subsequent candidate evidence node are multiplied to obtain the concatenation credibility between the preceding and subsequent candidate evidence nodes.

[0020] Optionally, the step of constructing a candidate evidence chain based on the numerical order of the center time of the candidate evidence nodes and the credibility of splicing between candidate evidence nodes, calculating the candidate evidence chain score, and selecting the optimal abnormal evidence chain specifically includes:

[0021] For each vehicle in the chain-constructed vehicle set, candidate evidence node pairs are established according to the sequence number of the candidate evidence nodes, ensuring that the sequence number of the preceding candidate evidence node in each pair is less than that of the following candidate evidence node. Candidate evidence node pairs are then filtered: if the center time value of the following candidate evidence node is greater than that of the preceding candidate evidence node, and the concatenation credibility between the preceding and following candidate evidence nodes is greater than zero, the candidate evidence node pair is included in the set of possible concatenation candidate evidence node pairs. For each candidate evidence node, the anomaly strength and identity of the candidate evidence node are compared. The credibility values ​​are multiplied to obtain the contribution value of the candidate evidence nodes; among the candidate evidence nodes of the same vehicle, they are arranged into an ordered candidate evidence chain in ascending order of their serial numbers; when an ordered candidate evidence chain contains one candidate evidence node, it is added to the candidate evidence chain set; when an ordered candidate evidence chain contains two or more candidate evidence nodes, the candidate evidence node pairs formed by adjacent chains are checked sequentially to see if they all belong to the set of concatenable candidate evidence node pairs, and the ordered candidate evidence chains that satisfy the check results are added to the candidate evidence chain set; for any candidate evidence chain, all candidates in the current candidate evidence chain are... The contribution values ​​of candidate evidence nodes are summed to obtain the total contribution of candidate evidence nodes. For candidate evidence chains containing two or more candidate evidence nodes, the concatenation credibility between candidate evidence nodes in adjacent chains is summed to obtain the total concatenation credibility. The total contribution of candidate evidence nodes is added to the total concatenation credibility to form the comprehensive value of the candidate evidence chain. The number of candidate evidence nodes in the candidate evidence chain is multiplied by two and then subtracted by one to form the normalized denominator of the candidate evidence chain. The comprehensive value of the candidate evidence chain is divided by the normalized denominator of the candidate evidence chain to obtain the candidate evidence chain score. Candidate evidence chains are selected from the set of candidate evidence chains. The candidate evidence chain with the highest evidence chain score is selected as the optimal abnormal evidence chain for the corresponding vehicle. When two or more candidate evidence chains have the same score, the candidate evidence chain with more candidate evidence nodes is selected. When candidate evidence chains have the same score and the same number of candidate evidence nodes, the candidate evidence chain with the earlier center time value of the first candidate evidence node is selected. When candidate evidence chains have the same score, the same number of candidate evidence nodes, and the same center time value of the first candidate evidence node, the candidate evidence chain is compared digit by digit starting from the first digit according to the candidate evidence node arrangement sequence, and the candidate evidence chain with the smallest candidate evidence node arrangement sequence number for the first time is selected.

[0022] Optionally, the step of determining abnormal vehicles based on the optimal abnormal evidence chain score, the average strength of vehicle evidence chains within the chain-constructed vehicle set, and the average deviation of vehicle evidence chain strength within the chain-constructed vehicle set, and outputting the abnormal vehicle detection result, specifically includes:

[0023] For each vehicle in the original vehicle object set, a vehicle evidence chain strength is generated. When the current vehicle belongs to the chain-constructed vehicle set, the score of the current vehicle's best abnormal evidence chain is used as the vehicle evidence chain strength. When a vehicle in the original vehicle object set does not belong to the chain-constructed vehicle set, the current vehicle's vehicle evidence chain strength is recorded as zero. The vehicle evidence chain strengths of all vehicles in the chain-constructed vehicle set are summed, and the sum is divided by the effective denominator of the chain-constructed vehicle set to obtain the average vehicle evidence chain strength in the chain-constructed vehicle set. For each vehicle in the chain-constructed vehicle set, the vehicle evidence chain strength of the current vehicle is subtracted from the vehicle evidence chain strength of the vehicles in the chain-constructed vehicle set. The average strength of the evidence chain is calculated, and the absolute value of the difference between the current vehicle's evidence chain strength and the average strength of the vehicle evidence chain strength within the chain-constructed vehicle set is taken to obtain the vehicle evidence chain strength deviation. The vehicle evidence chain strength deviations of all vehicles within the chain-constructed vehicle set are summed, and the sum is divided by the effective denominator of the chain-constructed vehicle set to obtain the average deviation of the vehicle evidence chain strength within the chain-constructed vehicle set. The average strength of the vehicle evidence chain strength within the chain-constructed vehicle set is added to the average deviation of the vehicle evidence chain strength within the chain-constructed vehicle set to form the abnormal vehicle determination threshold. An abnormal vehicle label is determined for each vehicle; if the current vehicle belongs to the chain-constructed vehicle set... If the strength of the vehicle's evidence chain is greater than zero and greater than or equal to the abnormal vehicle determination threshold, then the current vehicle is recorded as an abnormal vehicle. If the current vehicle does not belong to the set of vehicles in the chain construction, the strength of the current vehicle's evidence chain is equal to zero, or the strength of the current vehicle's evidence chain is less than the abnormal vehicle determination threshold, then the current vehicle is recorded as a non-abnormal vehicle. For vehicles with the abnormal vehicle label "abnormal vehicle," output the vehicle number, abnormal vehicle label, vehicle evidence chain strength, candidate evidence node sequence in the optimal abnormal evidence chain, and candidate... Select the anomaly type, anomaly strength, and time interval of the candidate evidence nodes; for each candidate evidence node in the optimal anomaly evidence chain, extract all trajectory records from the start trajectory record number to the end trajectory record number of the current trajectory segment; compare the extracted trajectory records to form the minimum longitude, maximum longitude, minimum latitude, and maximum latitude, and form a latitude and longitude range quadruple in the order of minimum longitude, maximum longitude, minimum latitude, and maximum latitude; output the latitude and longitude range quadruple for each candidate evidence node in the optimal anomaly evidence chain.

[0024] The present invention has the following beneficial effects:

[0025] 1. The original vehicle trajectory records are first converted into standardized vehicle trajectory objects, and the collection time, latitude and longitude, and vehicle object identification confidence values ​​are uniformly organized to ensure that subsequent anomaly analysis is based on stable data. Organizing trajectory records by collection time avoids the confusion of trajectory directions caused by out-of-order data; retaining valid vehicle objects eliminates vehicles with insufficient trajectory records or those unable to form a movement trend, reducing invalid calculations; and using vehicle object identification confidence values ​​in trajectory record filtering reduces the risk of false associations when multiple trajectory records exist at the same time.

[0026] 2. First, calculate the time interval, ground distance, and traffic speed between adjacent trajectory records, then divide the trajectory into segments based on the number of trajectory records. Its advantage lies in its ability to decompose long-term, long-distance, and multi-state vehicle trajectories into several segments with local motion characteristics, transforming anomaly detection from overall vehicle assessment to local segment identification. This scheme, through segmentation, allows each local trajectory interval to be analyzed independently, enhancing the ability to capture short-term and phased anomalies. Compared to existing technologies that rely on average vehicle speed, total distance, or fixed time windows to determine anomalies, this scheme can form segments based on the actual number of trajectory records, taking into account the differences in trajectory length between different vehicles, avoiding the problems of anomalies being masked by excessively long fixed windows and fragmented results due to excessively short fixed windows.

[0027] 3. A multi-dimensional anomaly evaluation is constructed for each trajectory segment, rather than relying solely on a single velocity threshold or a single positional offset to determine anomalies. Specifically, distance compression reflects the difference between the cumulative trajectory distance and the direct distance to the endpoint; low-speed congestion reflects the concentration of low-speed motion within the segment; and time void reflects the abnormal expansion of the sampling time interval. Furthermore, the anomaly type is recorded based on the dominant anomaly quantity. Different forms of anomalous behavior are decomposed into distinguishable anomaly sources, enabling the system not only to determine whether an anomaly exists but also to explain its main manifestations. Compared to existing technologies that uniformly categorize anomalies into fuzzy labels, this scheme, through anomaly quantity combinations and dominant type recording, gives subsequent candidate evidence nodes clear anomaly attributes, facilitating consistency comparison of anomaly types among different nodes in the evidence chain.

[0028] 4. Activated trajectory segments are screened using the mean intensity of anomalies within the vehicle's internal segments and the average deviation of segment anomaly intensity. These activated segments are then further organized into candidate evidence nodes. This approach does not include all segments with anomaly intensities greater than zero in the evidence chain; instead, it selects more representative anomalous segments based on the overall anomaly level of the vehicle's own trajectory segments. This reduces the interference of occasional noise, minor fluctuations, or changes in ordinary traffic conditions on the detection results, allowing candidate evidence nodes to more effectively represent significant anomalies in the vehicle's trajectory. Candidate evidence nodes also incorporate information such as time interval, center time, center latitude and longitude, anomaly type, anomaly intensity, and identity credibility, giving them spatial, temporal, and credibility attributes. Compared to existing technologies that directly use anomaly points or segments as judgment results, the candidate evidence nodes generated by this scheme are more suitable for subsequent connection, scoring, and output. They can transform scattered trajectory anomalies into structured evidence units, improving the interpretability and evidence organization capabilities of anomalous vehicle detection.

[0029] 5. Before constructing the evidence chain, the time reference, distance reference, and achievable traffic speed between candidate evidence nodes are calculated for vehicles containing candidate evidence nodes. This provides a reference scale for subsequent node splicing that is adapted to the vehicle itself. This avoids using a uniform fixed threshold to judge the connectability of nodes for all vehicles, as the trajectory sampling frequency, traffic speed, activity range, and trajectory density of different vehicles may vary significantly. By establishing a reference quantity based on the current vehicle's candidate evidence node distribution and traffic speed, the splicing judgment can better conform to the vehicle's own motion characteristics. Compared to existing technologies that directly connect anomalous segments at fixed time intervals or fixed spatial distances, this scheme provides an adaptive basis for evidence chain connection, improving the accuracy of cross-segment anomalous evidence integration and reducing judgment bias caused by differences in motion scales between different vehicles.

[0030] 6. The credibility of the splicing between preceding and subsequent candidate evidence nodes is decomposed into a combination of temporal continuity, spatial reachability, anomaly type consistency, and identity credibility. The judgment of whether two anomaly nodes constitute the same anomaly process is expanded from a simple temporal sequence to a multi-factor comprehensive evaluation. Temporal continuity measures whether the two nodes have a reasonable temporal connection; spatial reachability measures whether a vehicle can reach another anomaly location within a corresponding time frame; anomaly type consistency measures whether the anomaly behavior has continuity; and identity credibility reduces the impact of false associations of vehicle objects on the evidence chain. Compared with the existing method of simply concatenating anomaly fragments, this scheme assigns an interpretable splicing credibility to each pair of candidate evidence nodes, making the evidence chain construction more robust and providing clear technical basis for subsequent scoring.

[0031] 7. Multiple candidate evidence nodes are constructed into candidate evidence chains according to chronological order and concatenation credibility. An evidence chain score is formed by combining the contribution value of each candidate evidence node with the concatenation credibility. The optimal anomaly evidence chain is then selected from the candidate evidence chain set. Anomaly vehicles are no longer judged based on a single highest-ranking anomaly fragment, but rather by a complete anomaly process formed by a group of mutually supporting candidate evidence nodes. The candidate evidence chain score considers the anomaly strength of each node, its identity credibility, and the connection relationships between nodes, ensuring that the final selected evidence chain reflects both the degree of anomaly and the continuity of the anomaly process. Compared to existing technologies that output single anomaly points or fragments, this approach forms a more complete anomaly chain, reduces false alarms caused by isolated noise, and provides more organized results for subsequent anomaly vehicle interpretation, trajectory playback, and evidence presentation.

[0032] 8. This method uses the optimal anomaly evidence chain scoring system, combined with the average strength and average deviation of vehicle evidence chains within the vehicle set to determine anomaly vehicles. It outputs a quadruple of anomaly vehicle label, vehicle evidence chain strength, candidate evidence node sequence, anomaly type, anomaly strength, time interval, and latitude / longitude range. This achieves a closed loop from anomaly detection to evidence expression: it not only determines whether a vehicle is abnormal but also outputs the evidence chain structure, anomaly occurrence time, anomaly spatial range, and anomaly type information corresponding to the abnormal vehicle. Compared to existing technologies that only provide anomaly scores or alarms, this solution outputs more complete information, directly serving anomaly vehicle screening, law enforcement evidence collection, trajectory backtracking, and manual review, thus improving the credibility, traceability, and practical application value of the detection results. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the process of the present invention.

[0034] Figure 2 This is a schematic diagram illustrating the establishment of vehicle trajectory objects and the standardization of latitude and longitude trajectory recording in this invention.

[0035] Figure 3 This is a schematic diagram of the trajectory segment anomalies and candidate evidence nodes of the present invention.

[0036] Figure 4 This is a schematic diagram illustrating the evidence node splicing benchmark and splicing credibility of the present invention. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] Example, refer to Figure 1 A method for abnormal vehicle detection and evidence chain construction based on trajectory analysis, comprising:

[0039] Vehicle trajectory objects within the detection area are established. Latitude and longitude trajectory records are organized according to collection time, retaining valid vehicle objects and standardizing latitude and longitude, collection time, and vehicle object identification confidence values. The time interval, surface distance, and travel speed between adjacent trajectory records of valid vehicle objects are calculated, and trajectory segments are divided according to the number of trajectory records. The distance compression, low-speed dwell time, and time hole of each trajectory segment are calculated to obtain anomaly intensity, and the anomaly type corresponding to the dominant anomaly is recorded. Trajectory segments in an active state are selected based on the mean anomaly intensity of segments within the vehicle and the average deviation of segment anomaly intensity, generating candidate evidence nodes. For vehicles containing candidate evidence nodes, calculations are performed... Calculate the time reference, distance reference, and reachable traffic speed between candidate evidence nodes; for candidate evidence node pairs of the same vehicle, calculate the combined results of time continuity, spatial reachability, anomaly type consistency, and identity credibility to obtain the splicing credibility between candidate evidence nodes; construct candidate evidence chains according to the numerical order of the center time of candidate evidence nodes and the splicing credibility between candidate evidence nodes, calculate the candidate evidence chain score, and select the optimal abnormal evidence chain; determine abnormal vehicles based on the optimal abnormal evidence chain score, the average strength of vehicle evidence chains within the chain-constructed vehicle set, and the average deviation of vehicle evidence chain strength within the chain-constructed vehicle set, and output the abnormal vehicle detection results.

[0040] First, vehicle trajectory objects are established and their latitude, longitude, collection time, and vehicle object identification confidence values ​​are standardized. Then, the time interval, surface distance, and traffic speed between adjacent trajectory records are calculated, and the trajectory is divided into segments. Subsequently, the anomaly intensity is calculated based on the distance compression, low-speed dwell time, and time void, and active trajectory segments are selected to generate candidate evidence nodes. Further, node splicing confidence is formed based on the time reference, distance reference, achievable traffic speed, temporal continuity, spatial accessibility, anomaly type consistency, and identity credibility between nodes, constructing candidate evidence chains and selecting the optimal anomaly evidence chain. Finally, the abnormal vehicle is determined based on the statistical results of the vehicle evidence chain strength, and the detection result is output. This method solves the problems of false alarms, missed alarms, and insufficient interpretability in existing technologies that rely solely on single-point anomalies, single-segment speed anomalies, or fixed thresholds to determine vehicle anomalies. This method organizes trajectory anomalies from scattered fragments into evidence chains supported by temporal order, spatial accessibility, and identity credibility, which can more clearly explain the anomaly source, anomaly segment, anomaly type, and spatial range of the abnormal vehicle.

[0041] Reference Figure 2The process of establishing vehicle trajectory objects within the detection area, organizing latitude and longitude trajectory records according to the collection time, retaining valid vehicle objects, and standardizing latitude and longitude, collection time, and vehicle object identification confidence values ​​specifically includes:

[0042] An initial vehicle object set is established for all vehicles within the detection area, and the total number of initial vehicles is recorded. The trajectory records for each vehicle are arranged in ascending order of acquisition time. Trajectory records output from the upstream trajectory aggregation platform and associated with vehicle object numbers by the vehicle object recognition processing unit are received. For each trajectory record, the vehicle number, longitude, latitude, acquisition time, and vehicle object recognition confidence value are extracted. The vehicle object recognition confidence value is recorded as the degree of matching confidence between the trajectory record and the corresponding vehicle number, and the vehicle object recognition confidence value is limited to a range of zero to one. When multiple trajectory records exist at the same acquisition time, the vehicle object recognition confidence values ​​of each trajectory record are compared, and the trajectory record with the highest vehicle object recognition confidence value is retained. When multiple trajectory records with the same vehicle object recognition confidence value exist at the same acquisition time, the original import sequence number of each trajectory record is compared, and the trajectory record with the smallest original import sequence number is retained. The retained trajectory records are renumbered in strictly ascending order of acquisition time. The number of renumbered trajectory records for each vehicle is counted, and vehicles with at least two renumbered trajectory records are included in the valid vehicle objects. For each vehicle in the set of valid vehicle objects, a trajectory record set is created, with the trajectory records arranged in strictly ascending order of acquisition time. For each trajectory record, the longitude is converted to radians, the latitude is converted to radians, and the acquisition time is converted to seconds. The vehicle object identification confidence value for each trajectory record is also recorded. For any two latitude and longitude points, the square of the sine of half the difference in radians between the two latitudes is first calculated, followed by the cosine of the first latitude, the cosine of the second latitude, and the radians of the two longitudes. The product of the square of half the difference in latitude arc length is used to form the auxiliary value of the surface arc distance. The square root of the auxiliary value of the surface arc distance is then taken and the arcsine of the two latitude arc length differences is added to the product of the square of the first latitude cosine value, the second latitude cosine value, and the square of the half sine of the difference in longitude arc length. The result is then multiplied by two and the average radius of the Earth to form the surface arc distance between the two latitude and longitude points, where the average radius of the Earth is taken as 6,371,000 meters. The length reference constant is set to one meter, and the time reference constant is set to one second.

[0043] The original vehicle object set is established for all vehicles within the detection area as follows:

[0044] ;in, For the original collection of vehicle objects; This represents the original total number of vehicles;

[0045] The trajectory records for each vehicle are sorted in ascending order by collection time;

[0046] Each trajectory record is output by the upstream trajectory aggregation platform and associated with the vehicle object number by the vehicle object recognition and processing unit;

[0047] Each trajectory record includes at least the vehicle number, longitude, latitude, collection time, and vehicle object identification confidence value;

[0048] The vehicle object recognition confidence value represents the degree of reliability in matching the trajectory record with the corresponding vehicle number, and its value range is [value range missing]. ;

[0049] When multiple trajectory records exist at the same collection time, the trajectory record with the highest vehicle object recognition confidence value is retained;

[0050] When vehicle object recognition confidence values ​​are the same, the trajectory record with the smallest original import sequence number is retained;

[0051] The retained trajectory records will be renumbered in strict ascending order according to the collection time;

[0052] The first The number of trajectory records after the vehicle is renumbered is denoted as Vehicles with at least two re-numbered trajectory records are included in the set of valid vehicle objects, resulting in:

[0053] ;in, A collection of valid vehicle objects; Index the vehicle numbers;

[0054] For the The set of trajectory records for each vehicle, arranged in strict ascending order of acquisition time, is as follows:

[0055] ;in, For the first A collection of vehicle trajectory records; For the first The first car Track records; Record the sequence number of the trajectory;

[0056] Convert the longitude in each trajectory record to radians and record it as follows: Convert the latitude of each trajectory record to radians and record it as . The acquisition time in each trajectory record is converted into seconds and recorded as follows: The vehicle object identification confidence value in each trajectory record is recorded as follows: ;in, , The first vehicle number The trajectory records the longitude and latitude; For the first vehicle number The data collected at the time of each trajectory record; For the first vehicle number The confidence value for vehicle object identification recorded in each trajectory;

[0057] For any two latitude and longitude points and The auxiliary function for surface arc distance is constructed as follows: ;in, This is an auxiliary function for the arc distance on the Earth's surface. , These are the longitude and latitude of the first latitude and longitude point, respectively. , These are the longitude and latitude of the second latitude and longitude point, respectively.

[0058] The latitude and longitude surface arc distance function is established as follows:

[0059] ;in, It is a function of latitude and longitude surface arc distance; Take 6,371,000 meters as the average radius of the Earth;

[0060] Set length reference constant rice;

[0061] Set time base constant Second.

[0062] The calculation of the time interval, surface distance, and traffic speed of adjacent trajectory records of valid vehicle objects, and the division of trajectory segments according to the number of trajectory records, specifically includes:

[0063] For each vehicle's adjacent trajectory records, the time interval between adjacent trajectory records is obtained by subtracting the acquisition time value of the previous trajectory record from the acquisition time value of the later trajectory record; for each vehicle's adjacent trajectory records, the surface distance between the previous and subsequent trajectory records is calculated according to the method of calculating the surface arc distance of latitude and longitude; for each vehicle's adjacent trajectory records, the surface distance between adjacent trajectory records is divided by the time interval between adjacent trajectory records to obtain the travel speed between adjacent trajectory records; the number of trajectory records for each vehicle is reduced by one to obtain the length of the first candidate segment; the preset minimum segment length is compared with the length of the first candidate segment, and the larger of the two is taken as the length of the second candidate segment; the first candidate... The segment length is compared with the second candidate segment length, and the smaller value is taken as the segment length of the current vehicle; the number of trajectory records is reduced by one and divided by the segment length, and the result is rounded up to obtain the total number of trajectory segments for the current vehicle; the starting trajectory record number of each trajectory segment is calculated according to the trajectory segment number and the segment length; the candidate ending trajectory record number of each trajectory segment is calculated according to the trajectory segment number and the segment length, and the candidate ending trajectory record number is compared with the number of trajectory records for the current vehicle, and the smaller value is taken as the ending trajectory record number of the current trajectory segment; the trajectory record is extracted according to the starting trajectory record number and the ending trajectory record number of each trajectory segment to form the trajectory segment of the corresponding vehicle.

[0064] For the The adjacent trajectory records of the vehicles are calculated, and the time interval is as follows:

[0065] ;in, For the first vehicle number Track record and the first The time interval between each trajectory record;

[0066] For the The adjacent trajectory records of vehicles are used to calculate the distance to the ground surface, specifically:

[0067] ;in, For the first vehicle number Track record and the first The surface distance between the track records;

[0068] For the The adjacent trajectory records of vehicles are used to calculate the traffic speed, specifically:

[0069] ;in, For the first vehicle number Track record and the first The passage rate between track records;

[0070] Set the first Car segment length for:

[0071] ;

[0072] Calculate the first The total number of trajectory segments of the vehicle is ;

[0073] For the The first car Each trajectory segment determines the start and end trajectory record sequence numbers, specifically:

[0074] , ;in, The sequence number of the trajectory segment; , The first The first car The starting and ending trajectory records of each trajectory segment;

[0075] according to and Extract trajectory records to form the first The first car A trajectory segment .

[0076] Reference Figure 3 The calculation of the path compression, low-speed retention, and time hole of each trajectory segment yields the anomaly intensity, and the anomaly type corresponding to the dominant anomaly quantity is recorded, specifically including:

[0077] For each trajectory segment of each vehicle, the surface distance between adjacent trajectory records within the current trajectory segment is accumulated to obtain the cumulative distance of the segment; for each trajectory segment of each vehicle, the surface arc distance between the endpoints of the starting and ending trajectory records of the current trajectory segment is calculated according to the calculation method of latitude and longitude surface arc distance; the endpoint surface arc distance is subtracted from the cumulative distance of the segment, and the difference between the cumulative distance of the segment and the endpoint surface arc distance is compared with zero, and the larger value is taken as the distance difference; the cumulative distance of the segment is added to the length reference constant to obtain the normalized denominator of the distance; the distance difference is divided by the normalized denominator of the distance to obtain the distance compression of the current trajectory segment; the same vehicle... All adjacent traffic speeds of vehicles are sorted in ascending order of numerical value, and the lower median traffic speed is selected. Adjacent traffic speeds within each trajectory segment are compared one by one. When an adjacent traffic speed is less than or equal to the lower median traffic speed, a low-speed count is recorded; when an adjacent traffic speed is greater than the lower median traffic speed, no low-speed count is added. The low-speed count is divided by the number of adjacent movements compared within the current trajectory segment to obtain the low-speed dwell time of the current trajectory segment. The time intervals between all adjacent trajectory records for the same vehicle are summed, and the sum is divided by the number of adjacent trajectory record time intervals to obtain the average of the adjacent trajectory record time intervals. The time intervals between adjacent trajectory records within each trajectory segment are... Each time interval is processed individually. The average time interval between adjacent trajectory records is subtracted from the average time interval between adjacent trajectory records. The difference between the two is compared with zero, and the larger value is taken as the time interval excess. The time interval between adjacent trajectory records is added to the time base constant to obtain the time normalized denominator. The time interval excess is divided by the time normalized denominator to obtain the time hole sub-quantity of a single adjacent trajectory record time interval. All time hole sub-quantities within the same trajectory segment are summed, and the sum is divided by the number of adjacent motion quantities involved in the calculation within the current trajectory segment to obtain the time hole quantity of the current trajectory segment. The path compression, low-speed stagnation, and time void of the same trajectory segment are added together, and the sum is divided by three to obtain the anomaly intensity of the current trajectory segment. When the path compression is greater than or equal to the low-speed stagnation and the path compression is greater than or equal to the time void, the anomaly type of the current trajectory segment is recorded as path compression-dominated anomaly. When the path compression is less than the low-speed stagnation and the low-speed stagnation is greater than or equal to the time void, the anomaly type of the current trajectory segment is recorded as low-speed stagnation-dominated anomaly. When the path compression is less than the low-speed stagnation and the low-speed stagnation is less than the time void, the anomaly type of the current trajectory segment is recorded as time void-dominated anomaly.

[0078] For the A clip of a car Calculate the cumulative distance of the segments, specifically:

[0079] ;in, For the first vehicle number The cumulative distance of each trajectory segment;

[0080] The surface arc distance at the endpoints of the segment is calculated as follows: ;in, For the first vehicle number The surface arc distance at the endpoints of a trajectory segment; , The first vehicle number The longitude and latitude of the initial trajectory record for each trajectory segment; , The first vehicle number The longitude and latitude of the end trajectory segment are recorded.

[0081] Calculate the first vehicle number Path compression of a trajectory segment Specifically:

[0082] ;

[0083] The first All vehicles Sort by numerical ascending order, and then sort the next... The traffic rate is denoted as the lower median traffic rate. ;in, For the first The lower median traffic speed of adjacent vehicle traffic speeds;

[0084] Calculate the first vehicle number Low-speed retention of each trajectory segment Specifically:

[0085] ;in, This is an indicator function; it outputs 1 when the condition inside the parentheses is true and 0 when the condition is false.

[0086] Calculate the first Mean time interval between adjacent trajectory records of vehicles Specifically:

[0087] ;

[0088] Construct the first vehicle number Time gap of a trajectory segment Specifically:

[0089] ;

[0090] Construct the first vehicle number Anomaly intensity of each trajectory segment Specifically:

[0091] ;

[0092] Construct the first vehicle number Anomaly types of trajectory segments Specifically:

[0093] ;in, This indicates that the distance compression is the dominant outlier. This indicates that low-speed retention is the dominant anomaly. This indicates that time voids are the dominant anomaly.

[0094] Reference Figure 3 The step of filtering active trajectory segments based on the mean anomaly intensity and the mean deviation of anomaly intensity within the vehicle to generate candidate evidence nodes specifically includes:

[0095] For each vehicle, sum the anomaly intensities of all segments and divide the sum by the total number of trajectory segments for that vehicle to obtain the mean anomaly intensity of segments within the vehicle. For each trajectory segment, subtract the mean anomaly intensity of segments within the vehicle from the anomaly intensity of the current trajectory segment, and take the absolute value of the difference between the anomaly intensity of the current trajectory segment and the mean anomaly intensity of segments within the vehicle to obtain the segment anomaly intensity deviation. Sum the anomaly intensity deviations of all segments for the same vehicle and divide the sum by the total number of trajectory segments for that vehicle to obtain the average anomaly intensity deviation of segments. Add the mean anomaly intensity of segments within the vehicle and the average anomaly intensity deviation of segments to form a candidate activation threshold. For each trajectory segment, compare the anomaly intensity with the candidate... The system compares activation thresholds. When the anomaly intensity of a trajectory segment is greater than zero and greater than or equal to the candidate activation threshold, the candidate activation marker for the current trajectory segment is recorded as active. When the anomaly intensity of a trajectory segment is equal to zero, the candidate activation marker for the current trajectory segment is recorded as inactive. When the anomaly intensity of a trajectory segment is greater than zero and less than the candidate activation threshold, the candidate activation marker for the current trajectory segment is recorded as inactive. For each trajectory segment, the acquisition time value of the starting trajectory record and the acquisition time value of the ending trajectory record are added together, and the result is divided by two to obtain the center time value of the current trajectory segment. For trajectory segments in the active state, all trajectories within the current trajectory segment are... The longitude records are summed, and the sum is divided by the number of trajectory records in the current trajectory segment to obtain the center longitude. For an active trajectory segment, the latitude records of all trajectory records in the current trajectory segment are summed, and the sum is divided by the number of trajectory records in the current trajectory segment to obtain the center latitude. For an active trajectory segment, the vehicle object identification confidence values ​​of all trajectory records in the current trajectory segment are summed, and the sum is divided by the number of trajectory records in the current trajectory segment to obtain the identity credibility. All active trajectory segments are sorted in ascending order of their center time values. When two active trajectory segments have the same center time value, they are sorted in ascending order of the starting trajectory record number of the active segment. For each active trajectory segment, candidate evidence nodes are generated. Each candidate evidence node includes its time interval, center time value, center longitude, center latitude, anomaly type, anomaly intensity, and identity credibility. The time interval of the candidate evidence node consists of the acquisition time values ​​of the starting trajectory record and the ending trajectory record of the current trajectory segment. The candidate evidence node sequence number and the original trajectory segment sequence number corresponding to each candidate evidence node are recorded according to the arrangement result. The number of candidate evidence nodes for each vehicle is recorded. When there are no active trajectory segments for the same vehicle, the number of candidate evidence nodes for the current vehicle is recorded as zero, and no candidate evidence node is generated.

[0096] For the The mean of the anomaly intensity across all segments of the vehicle was calculated, yielding:

[0097] ;in, For the first The mean anomaly intensity of the vehicle's interior segments; the mean deviation of the segment anomaly intensity is calculated as follows: ;in, For the first Average deviation of vehicle segment anomaly intensity;

[0098] For fragments Construct candidate activation markers, specifically as follows:

[0099] ;in, For the first vehicle number Candidate activation markers for each trajectory segment Representing fragments Activated as a candidate anomalous fragment, Representing fragments No candidate evidence nodes are generated;

[0100] Calculate the first vehicle number The center time value of each trajectory segment Specifically:

[0101] ;in, , The first vehicle number The acquisition time values ​​of the start and end trajectories of each trajectory segment;

[0102] The center latitude and longitude of the activated segment are calculated as follows:

[0103] ;in, , The first vehicle number The center longitude and latitude of each trajectory segment;

[0104] Construct the first vehicle number Credibility of the identity of each trajectory segment Specifically:

[0105] ;

[0106] All will be satisfied The activated fragments are arranged in ascending order of their center time;

[0107] When two activation segments have the same center time, they are sorted in ascending order according to the starting trajectory record number of the activation segments. The original fragment number of each activated fragment is denoted as . ;in, The candidate evidence node number; For the first vehicle number The original trajectory segment sequence number corresponding to each candidate evidence node;

[0108] For the Each activated fragment generates candidate evidence nodes, specifically:

[0109] ;in, For the first vehicle number One candidate evidence node; Pick For the first vehicle number The time interval of each candidate evidence node; Pick For the first vehicle number The center time value of each candidate evidence node; Pick , Pick The first vehicle number The center longitude and latitude of each candidate evidence node; Pick For the first vehicle number Anomaly types of candidate evidence nodes; Pick For the first vehicle number The abnormal strength of each candidate evidence node; Pick For the first vehicle number The credibility of the identity of each candidate evidence node; , The first vehicle number Record the sequence number of the starting and ending trajectory of each original trajectory segment; , The first vehicle number The acquisition time values ​​of the start and end trajectories of each original trajectory segment; , , , , , The first vehicle number The center time value, center longitude, center latitude, anomaly type, anomaly intensity, and identity credibility of each original trajectory segment;

[0110] Record number The number of candidate evidence nodes for the vehicle is ;

[0111] When there is no satisfying When activating fragments, No candidate evidence nodes are generated.

[0112] Reference Figure 4 The step of calculating the time reference between candidate evidence nodes, the distance reference between candidate evidence nodes, and the achievable passage speed for vehicles containing candidate evidence nodes specifically includes:

[0113] Vehicles with at least one candidate evidence node are included in the chain-constructed vehicle set; the number of elements in the chain-constructed vehicle set is compared with one, and the larger value is used as the effective denominator for the chain-constructed vehicle set; when the chain-constructed vehicle set is empty, the summation result on the chain-constructed vehicle set is recorded as zero; for each vehicle in the chain-constructed vehicle set, all candidate evidence nodes are sorted in ascending order according to their center time values ​​to form a set of candidate evidence nodes for the same vehicle; for each vehicle in the chain-constructed vehicle set, the current vehicle's lower median traffic speed is subtracted from each adjacent traffic speed, and the absolute value of the difference between each adjacent traffic speed and the current vehicle's lower median traffic speed is taken to obtain the traffic speed deviation; the traffic speed deviations of the same vehicle are summed, and the summation result is divided by the number of adjacent traffic speeds of the current vehicle to obtain the average traffic speed deviation; the current vehicle's lower median traffic speed is added to the average traffic speed deviation to form the achievable traffic speed screening upper limit; adjacent traffic speeds not greater than the achievable traffic speed screening upper limit are selected from all adjacent traffic speeds of the current vehicle, and... The maximum value in the screening results is taken as the achievable passage speed of the current vehicle. For vehicles with more than one candidate evidence node in the chain-constructed vehicle set, the difference in the center time values ​​of adjacent candidate evidence nodes is calculated one by one, and all center time value differences are summed. The sum of the center time value differences of adjacent candidate evidence nodes is divided by the number of candidate evidence nodes of the current vehicle to obtain the time reference between the candidate evidence nodes of the current vehicle. When the number of candidate evidence nodes of the same vehicle is one, the sum of the center time value differences of adjacent candidate evidence nodes is recorded as zero. For vehicles with more than one candidate evidence node in the chain-constructed vehicle set, the center position arc distance of adjacent candidate evidence nodes is calculated one by one according to the calculation method of latitude and longitude surface arc distance, and all center position arc distances are summed. The sum of the center position arc distances of adjacent candidate evidence nodes is divided by the number of candidate evidence nodes of the current vehicle to obtain the distance reference between the candidate evidence nodes of the current vehicle. When the number of candidate evidence nodes of the same vehicle is one, the sum of the center position arc distances of adjacent candidate evidence nodes is recorded as zero.

[0114] Vehicles with at least one candidate evidence node are included in the chain to construct a vehicle set, resulting in: ;in, Build a collection of vehicles for the chain;

[0115] Chain construction of vehicle effective denominator Take the larger of the number of elements in the vehicle set constructed by the chain and 1;

[0116] When the set of vehicles constructed by the chain is empty, the summation result on the set of vehicles constructed by the chain is 0;

[0117] right For the vehicle in question, all candidate evidence nodes are arranged in ascending order of center time to form a set of candidate evidence nodes for the same vehicle, resulting in: ;in, For the first The set of candidate evidence nodes for the same vehicle;

[0118] right For vehicles, calculate the average deviation of traffic speed, specifically:

[0119] ;in, For the first Average deviation of vehicle traffic speed;

[0120] Calculate the first Vehicle achievable traffic speed Specifically:

[0121] ;

[0122] right The benchmark for calculating the time interval between adjacent candidate evidence nodes is:

[0123] ;in, For the first Time reference between vehicle nodes; The index of the benchmark summation between nodes; For the first vehicle number The center time value of each candidate evidence node;

[0124] when When the sum of adjacent candidate evidence nodes is zero, the result is 0.

[0125] right For adjacent candidate evidence nodes, the arc distance between the center positions of the nodes is calculated as follows:

[0126] ;in, For the first Vehicle node distance benchmark; , The first vehicle number The center latitude and longitude of each candidate evidence node;

[0127] when When the sum of adjacent candidate evidence nodes is 0, the result is 0.

[0128] Reference Figure 4The calculation of the temporal continuity, spatial reachability, anomaly type consistency, and identity credibility of candidate evidence nodes for the same vehicle yields the concatenation credibility between candidate evidence nodes, specifically including:

[0129] For each vehicle in the chain-constructed vehicle set, select preceding and subsequent candidate evidence nodes in order of their arrangement numbers, ensuring that the arrangement number of the preceding candidate evidence node is less than that of the subsequent candidate evidence node. Subtract the center time value of the preceding candidate evidence node from the center time value of the subsequent candidate evidence node to obtain the difference in center time values ​​between them. Divide this difference in center time values ​​by the sum of the time base and the time base constant between the candidate evidence nodes of the current vehicle to form the result. The time continuity quotient is calculated by taking the negative of the time continuity quotient and then performing a natural exponential operation to obtain the time continuity between preceding and subsequent candidate evidence nodes. Following the method for calculating latitude and longitude surface arc distances, the center position arc distance between the center longitude and center latitude of the preceding and subsequent candidate evidence nodes is calculated. The current achievable traffic speed of the vehicle is multiplied by the difference in center time values ​​between the preceding and subsequent candidate evidence nodes to form the reachable distance between candidate evidence nodes. The reachable distance between candidate evidence nodes is subtracted from the center position arc distance, and the center position arc distance is then calculated. The difference between the position arc distance and the reachable distance between candidate evidence nodes is compared with zero, and the larger value is taken as the excess reachable distance. The excess reachable distance is divided by the sum of the distance benchmark and length benchmark constant between candidate evidence nodes of the current vehicle to form the spatial reachability quotient. The spatial reachability quotient is then inversely calculated and subjected to natural exponentiation to obtain the spatial reachability between the preceding and subsequent candidate evidence nodes. The anomaly types of the preceding and subsequent candidate evidence nodes are compared. When they are the same, the anomaly type consistency is recorded as one. The anomaly types of the preceding and subsequent candidate evidence nodes are compared. When the anomaly type of the normal type differs from that of the subsequent candidate evidence node, the anomaly type of the preceding candidate evidence node is subtracted from that of the subsequent candidate evidence node, and the absolute value of the difference is taken. The absolute value of the difference is added to one, and the reciprocal of the sum is taken to obtain the anomaly type consistency between the preceding and subsequent candidate evidence nodes. The temporal continuity, spatial reachability, anomaly type consistency, identity credibility of the preceding candidate evidence node, and identity credibility of the subsequent candidate evidence node are multiplied to obtain the concatenation credibility between the preceding and subsequent candidate evidence nodes.

[0130] right And satisfy For the candidate evidence node pairs, execute steps S601 to S605, specifically as follows:

[0131] S601. Calculate the time continuity, specifically: ;in, , These are the preceding node number and the following node number in the candidate evidence node comparison sequence, respectively. For the first vehicle number The candidate evidence node and the first Temporal continuity between candidate evidence nodes; It is a natural exponential function;

[0132] S602, The arc distance at the center position is:

[0133] ;in, , The first vehicle number The center longitude and latitude of each candidate evidence node; , The first vehicle number The center longitude and latitude of each candidate evidence node; For the first vehicle number The candidate evidence node and the first The arc distance between the center positions of each candidate evidence node;

[0134] S603, the spatial reachability is:

[0135] ;

[0136] in, For the first vehicle number The candidate evidence node and the first Spatial reachability between candidate evidence nodes;

[0137] S604, The consistency of exception types is:

[0138] ;in, For the first vehicle number The candidate evidence node and the first Consistency of anomaly types among candidate evidence nodes;

[0139] S605, The credibility of the constructed splicing is:

[0140] ;in, For the first vehicle number The candidate evidence node and the first Credibility of node splicing between candidate evidence nodes; , The first vehicle number The, the The credibility of the identity of each candidate evidence node.

[0141] The process of constructing a candidate evidence chain based on the numerical order of the center times of candidate evidence nodes and the credibility of splicing candidate evidence nodes, calculating the candidate evidence chain score, and selecting the optimal abnormal evidence chain specifically includes:

[0142] For each vehicle in the chain-constructed vehicle set, candidate evidence node pairs are established according to the sequence number of the candidate evidence nodes, ensuring that the sequence number of the preceding candidate evidence node in each pair is less than that of the following candidate evidence node. Candidate evidence node pairs are then filtered: if the center time value of the following candidate evidence node is greater than that of the preceding candidate evidence node, and the concatenation credibility between the preceding and following candidate evidence nodes is greater than zero, the candidate evidence node pair is included in the set of possible concatenation candidate evidence node pairs. For each candidate evidence node, the anomaly strength and identity of the candidate evidence node are compared. The credibility values ​​are multiplied to obtain the contribution value of the candidate evidence nodes; among the candidate evidence nodes of the same vehicle, they are arranged into an ordered candidate evidence chain in ascending order of their serial numbers; when an ordered candidate evidence chain contains one candidate evidence node, it is added to the candidate evidence chain set; when an ordered candidate evidence chain contains two or more candidate evidence nodes, the candidate evidence node pairs formed by adjacent chains are checked sequentially to see if they all belong to the set of concatenable candidate evidence node pairs, and the ordered candidate evidence chains that satisfy the check results are added to the candidate evidence chain set; for any candidate evidence chain, all candidates in the current candidate evidence chain are... The contribution values ​​of candidate evidence nodes are summed to obtain the total contribution of candidate evidence nodes. For candidate evidence chains containing two or more candidate evidence nodes, the concatenation credibility between candidate evidence nodes in adjacent chains is summed to obtain the total concatenation credibility. The total contribution of candidate evidence nodes is added to the total concatenation credibility to form the comprehensive value of the candidate evidence chain. The number of candidate evidence nodes in the candidate evidence chain is multiplied by two and then subtracted by one to form the normalized denominator of the candidate evidence chain. The comprehensive value of the candidate evidence chain is divided by the normalized denominator of the candidate evidence chain to obtain the candidate evidence chain score. Candidate evidence chains are selected from the set of candidate evidence chains. The candidate evidence chain with the highest evidence chain score is selected as the optimal abnormal evidence chain for the corresponding vehicle. When two or more candidate evidence chains have the same score, the candidate evidence chain with more candidate evidence nodes is selected. When candidate evidence chains have the same score and the same number of candidate evidence nodes, the candidate evidence chain with the earlier center time value of the first candidate evidence node is selected. When candidate evidence chains have the same score, the same number of candidate evidence nodes, and the same center time value of the first candidate evidence node, the candidate evidence chain is compared digit by digit starting from the first digit according to the candidate evidence node arrangement sequence, and the candidate evidence chain with the smallest candidate evidence node arrangement sequence number for the first time is selected.

[0143] right For the vehicles, establish a set of connectable node pairs, specifically as follows:

[0144] ;in, For the first The set of connectable node pairs for vehicles; An ordered binary tuple indicates that the connection direction of the candidate evidence nodes starts from the first... The node points to the first One node;

[0145] For candidate evidence nodes The contribution value of the computing node is: ;in, For the first vehicle number The contribution value of each candidate evidence node in the evidence chain;

[0146] Will satisfy , , and when Time All meet The ordered candidate evidence chains constitute the candidate evidence chain set. ;in, For the first An ordered chain of candidate evidence for the vehicle; For the first A set of candidate evidence chains for the vehicle; For candidate chain of evidence The number of nodes included; The first in the candidate chain of evidence The candidate evidence node sequence number corresponding to each position within the chain; This refers to the position number of a node within the candidate evidence chain;

[0147] For any candidate chain of evidence Construct a candidate evidence chain score, specifically as follows:

[0148] ;in, For candidate chain of evidence The candidate evidence chain score;

[0149] Construct the first The optimal chain of evidence for the vehicle is ;

[0150] When two or more candidate evidence chains have the same score, the candidate evidence chain with more nodes is selected.

[0151] When candidate evidence chains have the same score and the same number of nodes, the candidate evidence chain with the earlier node center time is selected.

[0152] When candidate evidence chains have the same score, the same number of nodes, and the same center time of the first node, the candidate evidence chain with the smaller lexicographical order of the node sequence number is selected.

[0153] The lexicographical order of node numbers is determined by comparing each node in the candidate evidence chain sequentially, starting from the first node. The candidate evidence chain that first appears with a smaller node number is the candidate evidence chain with the smaller lexicographical order of node numbers.

[0154] The process of determining abnormal vehicles based on the optimal abnormal evidence chain score, the mean strength of vehicle evidence chains within the chain-constructed vehicle set, and the average deviation of vehicle evidence chain strength within the chain-constructed vehicle set, and outputting the abnormal vehicle detection result, specifically includes:

[0155] For each vehicle in the original vehicle object set, a vehicle evidence chain strength is generated. When the current vehicle belongs to the chain-constructed vehicle set, the score of the current vehicle's best abnormal evidence chain is used as the vehicle evidence chain strength. When a vehicle in the original vehicle object set does not belong to the chain-constructed vehicle set, the current vehicle's vehicle evidence chain strength is recorded as zero. The vehicle evidence chain strengths of all vehicles in the chain-constructed vehicle set are summed, and the sum is divided by the effective denominator of the chain-constructed vehicle set to obtain the average vehicle evidence chain strength in the chain-constructed vehicle set. For each vehicle in the chain-constructed vehicle set, the vehicle evidence chain strength of the current vehicle is subtracted from the vehicle evidence chain strength of the vehicles in the chain-constructed vehicle set. The average strength of the evidence chain is calculated, and the absolute value of the difference between the current vehicle's evidence chain strength and the average strength of the vehicle evidence chain strength within the chain-constructed vehicle set is taken to obtain the vehicle evidence chain strength deviation. The vehicle evidence chain strength deviations of all vehicles within the chain-constructed vehicle set are summed, and the sum is divided by the effective denominator of the chain-constructed vehicle set to obtain the average deviation of the vehicle evidence chain strength within the chain-constructed vehicle set. The average strength of the vehicle evidence chain strength within the chain-constructed vehicle set is added to the average deviation of the vehicle evidence chain strength within the chain-constructed vehicle set to form the abnormal vehicle determination threshold. An abnormal vehicle label is determined for each vehicle; if the current vehicle belongs to the chain-constructed vehicle set... If the strength of the vehicle's evidence chain is greater than zero and greater than or equal to the abnormal vehicle determination threshold, then the current vehicle is recorded as an abnormal vehicle. If the current vehicle does not belong to the set of vehicles in the chain construction, the strength of the current vehicle's evidence chain is equal to zero, or the strength of the current vehicle's evidence chain is less than the abnormal vehicle determination threshold, then the current vehicle is recorded as a non-abnormal vehicle. For vehicles with the abnormal vehicle label "abnormal vehicle," output the vehicle number, abnormal vehicle label, vehicle evidence chain strength, candidate evidence node sequence in the optimal abnormal evidence chain, and candidate... Select the anomaly type, anomaly strength, and time interval of the candidate evidence nodes; for each candidate evidence node in the optimal anomaly evidence chain, extract all trajectory records from the start trajectory record number to the end trajectory record number of the current trajectory segment; compare the extracted trajectory records to form the minimum longitude, maximum longitude, minimum latitude, and maximum latitude, and form a latitude and longitude range quadruple in the order of minimum longitude, maximum longitude, minimum latitude, and maximum latitude; output the latitude and longitude range quadruple for each candidate evidence node in the optimal anomaly evidence chain.

[0156] For the first vehicle object in the original vehicle object set The strength of the evidence chain for the vehicle is calculated as follows:

[0157] ;in, For the first The strength of the vehicle evidence chain; For the first The score for the optimal chain of evidence for the vehicle; The set difference represents the vehicles in the original set of vehicle objects that do not belong to the set of vehicles constructed by the chain;

[0158] The average strength of the evidence chain for all vehicles in the vehicle set constructed by the chain is:

[0159] ;in, Construct the average strength of the vehicle evidence chain within the vehicle set for the chain;

[0160] The average deviation of the vehicle evidence chain strength within the vehicle set is calculated using the chain of evidence construction.

[0161] ;in, Construct the average deviation of the vehicle evidence chain strength within the vehicle set for the chain;

[0162] Construct the first Abnormal vehicle tags Specifically:

[0163] ;in, Indicates the first The vehicle output is an abnormal vehicle; Indicates the first The vehicle is not output as an abnormal vehicle;

[0164] To satisfy For each vehicle, output the vehicle number, abnormal vehicle label, evidence chain strength, optimal evidence chain node sequence, evidence node anomaly type, evidence node anomaly strength, evidence node time interval, and a quadruple of latitude and longitude range of the trajectory segment corresponding to each evidence node in the optimal evidence chain.

[0165] Evidence node latitude and longitude range quadruple The four components are determined sequentially as follows: [The following text appears to be incomplete and requires further context for accurate translation.] Take from the trajectory record minimum value maximum value minimum value The maximum value; where, For the first vehicle number The latitude and longitude range of each evidence node is a quadruple.

[0166] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0167] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A trajectory analysis based abnormal vehicle detection and evidence chain construction method, characterized in that, include: Establish vehicle trajectory objects within the area to be detected, organize latitude and longitude trajectory records according to collection time, retain valid vehicle objects, and standardize latitude and longitude, collection time, and vehicle object identification confidence values; Calculate the time interval, ground distance, and traffic speed of adjacent trajectory records for valid vehicle objects, and divide the trajectory into segments according to the number of trajectory records; Calculate the path compression, low-speed retention, and time hole for each trajectory segment to obtain the anomaly intensity, and record the anomaly type corresponding to the dominant anomaly quantity. Based on the mean anomaly intensity of segments inside the vehicle and the average deviation of segment anomaly intensity, trajectory segments in the active state are selected to generate candidate evidence nodes; For vehicles containing candidate evidence nodes, calculate the time reference between candidate evidence nodes, the distance reference between candidate evidence nodes, and the achievable passage speed; For candidate evidence node pairs of the same vehicle, the combined results of time continuity, spatial reachability, anomaly type consistency and identity credibility are calculated to obtain the splicing credibility between candidate evidence nodes; Candidate evidence chains are constructed based on the numerical order of the central time of the candidate evidence nodes and the credibility of splicing between candidate evidence nodes. The scores of the candidate evidence chains are calculated, and the optimal abnormal evidence chain is selected. Abnormal vehicles are identified based on the optimal abnormal evidence chain score, the mean strength of the vehicle evidence chain within the chain-constructed vehicle set, and the average deviation of the vehicle evidence chain strength within the chain-constructed vehicle set, and the abnormal vehicle detection results are output. 2.The trajectory analysis based abnormal vehicle detection and evidence chain construction method of claim 1, wherein, The process of establishing vehicle trajectory objects within the area to be detected, organizing latitude and longitude trajectory records according to the collection time, retaining valid vehicle objects, and standardizing latitude and longitude, collection time, and vehicle object identification confidence values ​​specifically includes: Create an original set of vehicle objects for all vehicles within the detection area and record the total number of original vehicles; The trajectory records for each vehicle are arranged in ascending order of collection time; Receive trajectory records output from the upstream trajectory aggregation platform and associated with vehicle object numbers by the vehicle object recognition processing unit; For each trajectory record, extract the vehicle number, longitude, latitude, collection time, and vehicle object identification confidence value; The vehicle object identification confidence value is recorded as the degree of confidence in the match between the trajectory record and the corresponding vehicle number, and the vehicle object identification confidence value is limited to the range of zero to one. When multiple trajectory records exist at the same acquisition time, compare the vehicle object identification confidence values ​​of each trajectory record and retain the trajectory record with the highest vehicle object identification confidence value; When multiple vehicle object recognition confidence values ​​exist at the same acquisition time, compare the original import sequence number of each trajectory record and retain the trajectory record with the smallest original import sequence number. The retained trajectory records will be renumbered in strict ascending order according to the collection time; Count the number of trajectory records for each vehicle after renumbering, and include vehicles with at least two renumbered trajectory records into the set of valid vehicle objects. For each vehicle in the set of valid vehicle objects, a trajectory record set is established, and the trajectory records in the trajectory record set are arranged in strict ascending order according to the collection time. The longitude in each trajectory record is converted to radians, the latitude in each trajectory record is converted to radians, the acquisition time in each trajectory record is converted to seconds, and the vehicle object recognition confidence value in each trajectory record is recorded. For any two latitude and longitude points, first calculate the square of half the difference in latitude arcs, then calculate the product of the first latitude cosine, the second latitude cosine, and the square of half the difference in longitude arcs. Finally, add the square of half the difference in latitude arcs to the product of the first latitude cosine, the second latitude cosine, and the square of half the difference in longitude arcs to form the auxiliary value of the surface arc distance. After taking the square root of the auxiliary value of the surface arc distance, perform an arcsine operation. Multiply the result by two and the average radius of the Earth in turn to form the surface arc distance between two latitude and longitude points, where the average radius of the Earth is taken as 6,371,000 meters. Set the length reference constant to one meter and the time reference constant to one second. 3.The trajectory analysis based abnormal vehicle detection and evidence chain construction method of claim 2, wherein, The calculation of the time interval, surface distance, and traffic speed of adjacent trajectory records of valid vehicle objects, and the division of trajectory segments according to the number of trajectory records, specifically includes: For each vehicle's adjacent trajectory records, the time interval between adjacent trajectory records is obtained by subtracting the acquisition time value of the previous trajectory record from the acquisition time value of the later trajectory record. For each vehicle's adjacent trajectory records, the surface distance between the previous trajectory record and the next trajectory record is calculated according to the method for calculating the surface arc distance of latitude and longitude. For each vehicle's adjacent trajectory records, the ground distance between adjacent trajectory records is divided by the time interval between adjacent trajectory records to obtain the travel speed between adjacent trajectory records; Subtract one from the number of trajectory records for each vehicle to obtain the length of the first candidate segment; The preset minimum segment length 4 is compared with the first candidate segment length, and the larger of the two values ​​is taken as the second candidate segment length. The length of the first candidate segment is compared with the length of the second candidate segment, and the smaller value between the two is taken as the segment length of the current vehicle. Subtract one from the number of trajectory records, divide by the segment length, and round up to get the total number of trajectory segments for the current vehicle. Calculate the starting trajectory record number for each trajectory segment based on the segment number and segment length; Calculate the candidate end trajectory record number for each trajectory segment based on the trajectory segment number and segment length, and compare the candidate end trajectory record number with the current vehicle trajectory record number, taking the smaller value as the end trajectory record number of the current trajectory segment. The trajectory records are extracted based on the starting and ending trajectory record numbers of each trajectory segment to form the trajectory segment corresponding to the vehicle.

4. The trajectory analysis based abnormal vehicle detection and evidence chain construction method according to claim 3, characterized in that, The calculation of path compression, low-speed retention, and time voids for each trajectory segment yields the anomaly intensity, and the anomaly type corresponding to the dominant anomaly quantity is recorded, specifically including: For each trajectory segment of each vehicle, the ground distance between adjacent trajectory records within the current trajectory segment is accumulated to obtain the cumulative distance of the segment; For each trajectory segment of each vehicle, the endpoint surface arc distance between the starting and ending trajectory records of the current trajectory segment is calculated according to the method of calculating the latitude and longitude surface arc distance. Subtract the endpoint surface arc distance from the cumulative distance of the segment, and compare the difference between the cumulative distance of the segment and the endpoint surface arc distance with zero, taking the larger value as the distance difference; The normalized denominator of the distance is obtained by adding the cumulative distance of the segments to the length reference constant; Divide the distance difference by the distance normalization denominator to obtain the distance compression of the current trajectory segment; Sort all adjacent traffic speeds of the same vehicle in ascending order of numerical value, and select the lower median traffic speed; For each trajectory segment, the adjacent traffic speeds are compared one by one. When the adjacent traffic speed is less than or equal to the lower median traffic speed, a low speed count is recorded. When the adjacent traffic speed is greater than the lower median traffic speed, no low speed count is added. Divide the low-speed count by the number of adjacent motion quantities participating in the comparison within the current trajectory segment to obtain the low-speed dwell time of the current trajectory segment. The time intervals between all adjacent trajectory records of the same vehicle are summed, and the sum is divided by the number of time intervals between adjacent trajectory records to obtain the average time interval between adjacent trajectory records. The time intervals of adjacent trajectory records within each trajectory segment are processed one by one. The average time interval of adjacent trajectory records is subtracted from the time interval of adjacent trajectory records. The difference between the time interval of adjacent trajectory records and the average time interval of adjacent trajectory records is compared with zero, and the larger value of the two is taken as the time interval excess. Add the time reference constant to the time interval between adjacent trajectory records to obtain the time normalization denominator, and divide the time interval excess by the time normalization denominator to obtain the time hole sub-quantity of the time interval between individual adjacent trajectory records. Summing up all time-drain subquantities within the same trajectory segment and dividing the sum by the number of adjacent motion quantities involved in the calculation within the current trajectory segment yields the time-drain quantity of the current trajectory segment. Add up the path compression, low-speed retention, and time void of the same trajectory segment, and divide the sum by three to obtain the anomaly intensity of the current trajectory segment. When the distance compression is greater than or equal to the low-speed congestion and the distance compression is greater than or equal to the time hole, the anomaly type of the current trajectory segment is recorded as the distance compression-dominated anomaly. When the path compression is less than the low-speed stagnation and the low-speed stagnation is greater than or equal to the time hole, the anomaly type of the current trajectory segment is recorded as the low-speed stagnation dominant anomaly. When the path compression is less than the low-speed congestion and the low-speed congestion is less than the time hole, the anomaly type of the current trajectory segment is recorded as the time hole dominant anomaly.

5. The trajectory analysis based abnormal vehicle detection and evidence chain construction method according to claim 4, characterized in that, The step of filtering active trajectory segments based on the mean anomaly intensity and the average deviation of anomaly intensity within the vehicle segment, and generating candidate evidence nodes, specifically includes: The summation of all segment anomaly intensities for each vehicle is then divided by the total number of trajectory segments for the current vehicle to obtain the average segment anomaly intensity within the vehicle. For each trajectory segment, the abnormal intensity of the current trajectory segment is subtracted from the average abnormal intensity of the segments inside the vehicle, and the absolute value of the difference between the abnormal intensity of the current trajectory segment and the average abnormal intensity of the segments inside the vehicle is taken to obtain the segment abnormal intensity deviation. The average deviation of anomaly intensity for all segments of the same vehicle is obtained by summing the summation and dividing the result by the total number of trajectory segments of the current vehicle. The mean of the anomaly intensity of the segments inside the vehicle is added to the average deviation of the anomaly intensity of the segments to form the candidate activation threshold; The anomaly intensity of each trajectory segment is compared with the candidate activation threshold. When the anomaly intensity of the trajectory segment is greater than zero and greater than or equal to the candidate activation threshold, the candidate activation mark of the current trajectory segment is recorded as the activation state. When the anomaly intensity of a trajectory segment is zero, the candidate activation flag of the current trajectory segment is recorded as inactive. When the anomaly intensity of a trajectory segment is greater than zero and less than the candidate activation threshold, the candidate activation flag of the current trajectory segment is recorded as inactive. For each trajectory segment, add the acquisition time value of the starting trajectory record of the current trajectory segment to the acquisition time value of the ending trajectory record, and divide the sum by two to obtain the center time value of the current trajectory segment. For an active trajectory segment, sum the longitudes of all trajectory records within the current trajectory segment, and divide the sum by the number of trajectory records within the current trajectory segment to obtain the center longitude; For an active trajectory segment, sum the latitudes of all trajectory records within the current trajectory segment, and divide the sum by the number of trajectory records within the current trajectory segment to obtain the center latitude. For an active trajectory segment, sum the vehicle object identification confidence values ​​of all trajectory records within the current trajectory segment, and divide the sum by the number of trajectory records within the current trajectory segment to obtain the identity credibility. Arrange all active trajectory segments in ascending order of their center time values; When the center time values ​​of two active trajectory segments are the same, they are arranged in ascending order according to the starting trajectory record number of the active segment; For each active trajectory segment, candidate evidence nodes are generated. The candidate evidence node includes the candidate evidence node time interval, center time value, center longitude, center latitude, anomaly type, anomaly intensity, and identity credibility. The candidate evidence node time interval is composed of the acquisition time value of the starting trajectory record and the acquisition time value of the ending trajectory record of the current trajectory segment. Record the candidate evidence node arrangement number and the original trajectory segment number corresponding to each candidate evidence node according to the arrangement results. Record the number of candidate evidence nodes for each vehicle; When there are no active trajectory segments for the same vehicle, the number of candidate evidence nodes for the current vehicle is recorded as zero, and no candidate evidence nodes are generated.

6. The method for abnormal vehicle detection and evidence chain construction based on trajectory analysis according to claim 5, characterized in that, For vehicles containing candidate evidence nodes, the calculation of the time reference between candidate evidence nodes, the distance reference between candidate evidence nodes, and the achievable traffic speed specifically includes: Vehicles with at least one candidate evidence node are included in the chain to construct a vehicle set. Compare the number of elements in the chain-built vehicle set with one, and take the larger of the two values ​​as the effective denominator for the chain-built vehicle set. When the set of vehicles constructed by the chain is empty, the summation result on the set of vehicles constructed by the chain is recorded as zero; For each vehicle in the chain-constructed vehicle set, all candidate evidence nodes are sorted in ascending order according to the center time value to form a set of candidate evidence nodes for the same vehicle. For each vehicle in the chain-constructed vehicle set, subtract the current vehicle's lower median traffic rate from each adjacent traffic rate, and take the absolute value of the difference between each adjacent traffic rate and the current vehicle's lower median traffic rate to obtain the traffic rate deviation. The average speed deviation is obtained by summing all the speed deviations of the same vehicle and dividing the sum by the number of adjacent speeds of the current vehicle. The lower median speed of the current vehicles is added to the average deviation of the speed to form the upper limit of the achievable speed screening. Filter the adjacent traffic speeds that are no greater than the upper limit of the achievable traffic speed from all adjacent traffic speeds of the current vehicle, and take the maximum value in the filtering results as the achievable traffic speed of the current vehicle. For vehicles in the chain-constructed vehicle set with more than one candidate evidence node, calculate the difference in center time values ​​between adjacent candidate evidence nodes one by one, and sum all the differences in center time values. The summation of the differences in the center time values ​​of adjacent candidate evidence nodes is divided by the number of candidate evidence nodes of the current vehicle to obtain the time reference between the candidate evidence nodes of the current vehicle. When there is only one candidate evidence node for the same vehicle, the sum of the differences in the center time values ​​of adjacent candidate evidence nodes is recorded as zero. For vehicles with more than one candidate evidence node in the chain-constructed vehicle set, calculate the center position arc distance of adjacent candidate evidence nodes one by one according to the calculation method of latitude and longitude surface arc distance, and sum all the center position arc distances; Divide the sum of the arc distances between the center positions of adjacent candidate evidence nodes by the number of candidate evidence nodes of the current vehicle to obtain the distance benchmark between the candidate evidence nodes of the current vehicle. When there is only one candidate evidence node for the same vehicle, the sum of the arc distances between the center positions of adjacent candidate evidence nodes is recorded as zero.

7. The method for abnormal vehicle detection and evidence chain construction based on trajectory analysis according to claim 6, characterized in that, The combined results of calculating the temporal continuity, spatial reachability, anomaly type consistency, and identity credibility of candidate evidence node pairs for the same vehicle are used to obtain the concatenation credibility between candidate evidence nodes, specifically including: For each vehicle in the chain-constructed vehicle set, select the preceding and following candidate evidence nodes according to the order of their arrangement numbers, ensuring that the arrangement number of the preceding candidate evidence node is less than the arrangement number of the following candidate evidence node. Subtract the center time value of the preceding candidate evidence node from the center time value of the subsequent candidate evidence node to obtain the difference in center time values ​​between the preceding and subsequent candidate evidence nodes. Divide the difference in center time values ​​between the preceding and subsequent candidate evidence nodes by the sum of the time base and the time base constant between the candidate evidence nodes of the current vehicle to form the time continuity quotient. Take the negative of the time continuity quotient and perform natural exponentiation to obtain the time continuity between the preceding and subsequent candidate evidence nodes. Based on the method for calculating the surface arc distance of latitude and longitude, calculate the center position arc distance between the center longitude and center latitude of the preceding candidate evidence node and the center longitude and center latitude of the subsequent candidate evidence node; Multiply the current vehicle's achievable speed by the difference in center time values ​​between the preceding and subsequent candidate evidence nodes to form the reachable distance between candidate evidence nodes; Subtract the reachable distance between candidate evidence nodes from the arc distance at the center position, and compare the difference between the arc distance at the center position and the reachable distance between candidate evidence nodes with zero. Take the larger of the two values ​​as the excess reachable distance. The spatial reachability quotient is formed by dividing the distance beyond the reachable distance by the sum of the distance benchmark and length benchmark constant between the candidate evidence nodes of the current vehicle. The spatial reachability quotient is then inversely calculated by natural exponentiation to obtain the spatial reachability between the preceding and subsequent candidate evidence nodes. Compare the anomaly types of preceding candidate evidence nodes with those of subsequent candidate evidence nodes. When they are the same, record the anomaly type consistency as one. When the anomaly type of the preceding candidate evidence node is different from the anomaly type of the following candidate evidence node, the anomaly type of the preceding candidate evidence node is subtracted from the anomaly type of the following candidate evidence node, and the absolute value of the difference in anomaly types is taken. Add the absolute value of the anomaly type difference to one, and take the reciprocal of the sum to obtain the anomaly type consistency between the preceding and subsequent candidate evidence nodes. Multiply the temporal continuity, spatial reachability, anomaly type consistency, identity credibility of the preceding candidate evidence node, and identity credibility of the following candidate evidence node to obtain the concatenation credibility between candidate evidence nodes between the preceding and following candidate evidence nodes.

8. The trajectory analysis based abnormal vehicle detection and evidence chain construction method according to claim 7, characterized in that, The process of constructing a candidate evidence chain based on the numerical order of the center times of candidate evidence nodes and the credibility of splicing candidate evidence nodes, calculating the candidate evidence chain score, and selecting the optimal abnormal evidence chain specifically includes: For each vehicle in the chain-constructed vehicle set, establish a pair of candidate evidence nodes according to the order of the candidate evidence node arrangement numbers, and make the order number of the preceding candidate evidence node in the candidate evidence node pair less than the order number of the following candidate evidence node. Candidate evidence node pairs are screened. When the center time value of the subsequent candidate evidence node is greater than the center time value of the preceding candidate evidence node and the splicing credibility between the preceding and subsequent candidate evidence nodes is greater than zero, the candidate evidence node pair is included in the set of splicable candidate evidence node pairs. For each candidate evidence node, the anomaly strength of the candidate evidence node is multiplied by the identity credibility to obtain the contribution value of the candidate evidence node; In the candidate evidence nodes of the same vehicle, an ordered candidate evidence chain is formed in ascending order of the candidate evidence node arrangement number; When an ordered candidate evidence chain contains a candidate evidence node, the ordered candidate evidence chain is included in the candidate evidence chain set. When an ordered candidate evidence chain contains two or more candidate evidence nodes, check in turn whether the candidate evidence node pairs composed of candidate evidence nodes in adjacent chains all belong to the set of splicable candidate evidence node pairs, and include the ordered candidate evidence chain that satisfies the check result into the candidate evidence chain set. For any candidate evidence chain, sum the contribution values ​​of all candidate evidence nodes in the current candidate evidence chain to obtain the total contribution of the candidate evidence nodes; For a candidate evidence chain containing two or more candidate evidence nodes, the concatenation credibility between candidate evidence nodes in adjacent chains is summed to obtain the total concatenation credibility. The sum of the contributions of the candidate evidence nodes is added to the sum of the splicing credibility to form the comprehensive value of the candidate evidence chain; Multiply the number of candidate evidence nodes contained in the candidate evidence chain by two and then subtract one to form the normalized denominator of the candidate evidence chain. Divide the comprehensive value of the candidate evidence chain by the normalized denominator of the candidate evidence chain to obtain the candidate evidence chain score. The candidate evidence chain with the highest score is selected from the candidate evidence chain set as the optimal abnormal evidence chain for the corresponding vehicle. When two or more candidate evidence chains have the same score, the candidate evidence chain with more candidate evidence nodes is selected. When candidate evidence chains have the same score and the same number of candidate evidence nodes, the candidate evidence chain with the earlier center time value of the first candidate evidence node is selected. When the candidate evidence chains have the same score, the same number of candidate evidence nodes, and the same center time value of the first candidate evidence node, the candidate evidence chains are compared digit by digit starting from the first digit according to the candidate evidence node arrangement sequence of the candidate evidence chains, and the candidate evidence chain with the smaller candidate evidence node arrangement sequence number appears for the first time is selected.

9. The trajectory analysis based abnormal vehicle detection and evidence chain construction method according to claim 8, characterized in that, The process of determining abnormal vehicles based on the optimal abnormal evidence chain score, the average strength of vehicle evidence chains within the chain-constructed vehicle set, and the average deviation of vehicle evidence chain strength within the chain-constructed vehicle set, and outputting the abnormal vehicle detection results, specifically includes: forming a vehicle evidence chain strength for each vehicle in the original vehicle object set; when the current vehicle belongs to the chain-constructed vehicle set, using the score of the current vehicle's optimal abnormal evidence chain as the vehicle evidence chain strength. When a vehicle in the original vehicle object set does not belong to the chain-constructed vehicle set, the vehicle evidence chain strength of the current vehicle is recorded as zero. Summing the vehicle evidence chain strengths of all vehicles in the chain-constructed vehicle set, and dividing the sum by the effective denominator of the chain-constructed vehicle set, yields the mean vehicle evidence chain strength within the chain-constructed vehicle set. For each vehicle in the chain-constructed vehicle set, subtract the average vehicle evidence chain strength in the chain-constructed vehicle set from the current vehicle's vehicle evidence chain strength, and take the absolute value of the difference between the current vehicle's vehicle evidence chain strength and the average vehicle evidence chain strength in the chain-constructed vehicle set to obtain the vehicle evidence chain strength deviation. The deviation of the vehicle evidence chain strength of all vehicles in the chain-constructed vehicle set is summed, and the summation result is divided by the effective denominator of the chain-constructed vehicle set to obtain the average deviation of the vehicle evidence chain strength in the chain-constructed vehicle set. The average strength of the vehicle evidence chain within the chain-constructed vehicle set is added to the average deviation of the vehicle evidence chain strength within the chain-constructed vehicle set to form the abnormal vehicle determination threshold. For each vehicle, an abnormal vehicle label is determined. When the current vehicle belongs to the chain-constructed vehicle set, the vehicle evidence chain strength of the current vehicle is greater than zero, and the vehicle evidence chain strength of the current vehicle is greater than or equal to the abnormal vehicle determination threshold, the abnormal vehicle label of the current vehicle is recorded as an abnormal vehicle. When the current vehicle does not belong to the chain-built vehicle set, the vehicle evidence chain strength of the current vehicle is equal to zero, or the vehicle evidence chain strength of the current vehicle is less than the abnormal vehicle judgment threshold, the abnormal vehicle label of the current vehicle is recorded as a non-abnormal vehicle. For vehicles labeled as abnormal vehicles, output the vehicle number, abnormal vehicle label, vehicle evidence chain strength, candidate evidence node sequence in the optimal abnormal evidence chain, abnormal type of candidate evidence node in the optimal abnormal evidence chain, abnormal strength of candidate evidence node, and time interval of candidate evidence node. For each candidate evidence node in the optimal abnormal evidence chain, extract all trajectory records between the start trajectory record number and the end trajectory record number of the current trajectory segment. In all the extracted trajectory records, the minimum longitude, maximum longitude, minimum latitude, and maximum latitude are compared to form a quadruple of longitude and latitude ranges, and these quadruples are formed in the order of minimum longitude, maximum longitude, minimum latitude, and maximum latitude. Output the latitude and longitude range quadruples of the trajectory segment corresponding to each candidate evidence node in the optimal abnormal evidence chain.

Images