A method and system for interactive vehicle role recognition in autonomous driving scenarios
By acquiring vehicle trajectory and map information, the interactive vehicle roles are filtered and determined, solving the misjudgment problem in the existing technology of interactive vehicle role recognition, and realizing accurate role recognition and decision support in complex scenarios.
Patent Information
- Application Number
- CN202611259214.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-19
- Publication Date
- 2026-11-17
AI Technical Summary
Existing technologies struggle to accurately identify the role of interactive vehicles in autonomous driving scenarios, especially when trajectory segments are incomplete, map matching is unstable, or the same continuous driving route is divided into multiple lane segments, which can easily lead to misjudgments. Furthermore, it is difficult to determine the active vehicle based solely on spatial proximity or vehicle sequence.
By acquiring vehicle trajectory information and road map information, candidate interactive vehicle pairs are screened, interaction severity is calculated and interaction time periods are divided, strong map judgment is made by combining map information and motion characteristics, active interactive vehicles and target vehicles are identified, and role determination is made when there is insufficient evidence using map initiative score or kinematic initiative score, and ambiguous scenarios are eliminated.
It improves the reliability and consistency of interactive vehicle role recognition, reduces misjudgments, supports behavior prediction and autonomous driving decision control, adapts to different road geometry conditions, and reduces the impact of incorrect labels on the model.
Smart Images

Figure CN123104689A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of autonomous driving scene understanding, vehicle behavior recognition and control decision technology, and specifically relates to an interactive vehicle role recognition method and system for autonomous driving entry scenarios. Background Technology
[0002] With the development of autonomous driving and intelligent connected vehicle technologies, vehicles need to accurately understand the interactive relationships between surrounding traffic participants in human-machine hybrid driving traffic environments. Vehicle actions such as cutting in, changing lanes, and merging on ramps are typically accompanied by changes in lateral position, competition for longitudinal clearance, and adjustments to traffic priority. In these interactions, the vehicle actively initiating lateral movement and the target vehicle it approaches or enters the driving lane perform different behavioral functions. Distinguishing between the two helps clarify the active and responsive relationships of vehicle behavior. Therefore, simply determining whether a cut-in has occurred is insufficient to meet the needs of behavior prediction, risk assessment, and control decision-making; further identification is required of the vehicles that primarily alter lateral relationships and their main interactive objects.
[0003] Existing methods often classify vehicle roles based on lane number changes, vehicle position (front and rear), final lane location, or vehicle order within the data structure, which fails to reflect true interactive initiative. When trajectory segments are incomplete, map matching is unstable, or the same continuous driving route is divided into multiple interconnected lane segments, relying solely on start and end lane markings can easily lead to misjudgments. Furthermore, when map information is insufficient, directly using global coordinates to determine lateral movement may be affected by road curvature and ramp geometry. Additionally, a decrease in the lateral distance between two vehicles only indicates lateral convergence and cannot directly determine which vehicle is the active driver.
[0004] In multi-vehicle scenarios, it is also necessary to select the key interactive vehicles most relevant to the intervention behavior from among multiple vehicles. Using only fixed distance or nearest neighbor rules may select ordinary nearby vehicles instead of the main conflicting vehicles. Furthermore, real-world data contains ambiguous situations such as simultaneous lane changes, proactive approach, and conflicting evidence; forcibly assigning roles can easily lead to incorrect role labeling. Therefore, there is an urgent need for an interactive vehicle role recognition method that can select key vehicles based on the degree of traffic conflict and comprehensively classify and determine roles based on road map information and vehicle motion characteristics. Summary of the Invention
[0005] The purpose of this invention is to provide a method and system for identifying interactive vehicle roles in autonomous driving entry scenarios, which can overcome the shortcomings of existing technologies such as the reliance on spatial proximity for key interactive vehicle screening, the reliance on vehicle sequence or final lane for entry role division, global lateral displacement distortion when the road is curved, and forced role assignment when there is insufficient evidence.
[0006] To achieve the above objectives, the solution of the present invention is:
[0007] A method for recognizing interactive vehicle roles in autonomous driving scenarios includes the following steps:
[0008] Step 1: Obtain vehicle trajectory information and road map information;
[0009] Step 2: Iterate through pairs of moving vehicles at the same time, predict short-term movement paths based on vehicle trajectory information, and filter to obtain candidate interactive vehicle pairs;
[0010] Step 3: Calculate the interaction severity of candidate vehicle pairs, divide the moments when the interaction severity threshold is met consecutively into interaction time periods, and determine the key vehicle pairs within the interaction time periods; in specific implementation, the moments when the interaction severity threshold is met consecutively are divided into interaction time periods; when the interaction time period involves only one vehicle pair, it is directly used as the key vehicle pair; when it involves multiple candidate vehicle pairs, the key vehicle pairs are determined according to the maximum interaction severity of each vehicle pair within the interaction time period.
[0011] Step 4: Take the intersection of the timestamps of the key interactive vehicles as the common observable time period, perform time alignment and lane matching on the trajectories of the two key interactive vehicles in the common observable time period, determine whether the vehicle map information is reliable, and construct the original route set, original driving corridor and adjacent lane set of each vehicle respectively.
[0012] Step 5: Based on the judgment results of whether the map information of each vehicle is reliable, and combined with the original route set, original driving corridor and adjacent lane set of each vehicle, a strong map judgment is made, and the active interactive vehicle and target vehicle are identified based on the judgment results.
[0013] Specifically, in step 5, the actively interacting vehicle and the target vehicle are identified based on the following determination results:
[0014] Case 1: If and only if one vehicle meets the strong map determination criteria, the vehicle that meets the strong map determination criteria will be identified as the active interaction vehicle, and the other vehicle will be identified as the target vehicle.
[0015] Case 2: If neither vehicle meets the criteria for a strong map, and the map information of at least one key interactive vehicle is reliable, the active interactive vehicle and the target vehicle are determined based on the map initiative scores of the two vehicles.
[0016] Case 3: If neither vehicle meets the strong map determination criteria and the map information of both key interactive vehicles is unreliable, the active interactive vehicle and the target vehicle are determined based on the kinematic initiative scores of the two vehicles.
[0017] Case 4: When both vehicles meet the strong map determination criteria, the scene is marked as an ambiguous scene, and the ambiguous scene is removed from the training samples that need to determine the active interaction vehicle.
[0018] Specifically, the method for determining the active interaction vehicle and the target vehicle based on the map initiative score in scenario 2 or the kinematic initiative score in scenario 3 is as follows:
[0019] The vehicle with the higher map initiative score or kinematic initiative score is identified as the active interaction vehicle, and the other vehicle is identified as the target vehicle, provided that the following conditions are met simultaneously:
[0020] , , , ,
[0021] in, The larger of the two vehicle initiative scores is taken as the initiative score, which can be either the map initiative score or the kinematic initiative score. The smaller of the two vehicle initiative scores is taken as the initiative score, which can be either the map initiative score or the kinematic initiative score. This indicates the minimum initiative score required to confirm the active interaction of vehicles. This represents the minimum absolute difference that the initiative scores of the two vehicles must satisfy. This represents the minimum threshold for the ratio of the two vehicles' initiative scores. This indicates the minimum lateral contribution percentage required for vehicles to achieve a higher initiative score. and Representing vehicles and vehicles Contribution to reducing the lateral distance between the two vehicles; This indicates the lateral approach contribution of vehicles with higher initiative scores, when the vehicle When the initiative score is high, When the vehicle When the initiative score is high, , This indicates a preset positive number to prevent the denominator from being zero.
[0022] Specifically, a scene will be marked as ambiguous if any of the following conditions occur:
[0023] Both key interactive vehicles meet the criteria for strong map detection; or,
[0024] In the map-based initiative scoring or kinematic initiative scoring phase, the role acceptance criteria of minimum initiative score, absolute difference in initiative scores, initiative score ratio, and lateral proximity contribution ratio were not simultaneously met; or,
[0025] The shared observable trajectories of the two key interactive vehicles are insufficient to support role determination.
[0026] When a scene is marked as ambiguous, the active interactive vehicle identifier and the target vehicle identifier are set to empty, the ambiguity flag is set to true, and the ambiguous scene is removed from the training samples that need to determine the active interactive vehicle.
[0027] In particular, for scenarios 1, 2 and 3 in which the active interactive vehicle and the target vehicle have been determined, the method further includes determining the interaction determination time and determining the longitudinal positional relationship between the active interactive vehicle and the target vehicle based on the interaction determination time.
[0028] In Case 1, the moment when the active interactive vehicle enters the original route of the target vehicle is determined as the interaction determination moment; in Cases 2 and 3, the moment when the lateral distance between the active interactive vehicle and the target vehicle first falls below a preset threshold is determined as the interaction determination moment.
[0029] Specifically, based on the determination of the interaction timing, the longitudinal positional relationship between the actively interacting vehicle and the target vehicle is determined, including:
[0030] If the interaction determination time has been determined, the following calculations are performed using the interaction determination time; if the interaction determination time has not been determined, the time when the lateral distance between the two vehicles is the smallest is used as the interaction determination time for the following calculations.
[0031] At the moment of interaction determination, the longitudinal position difference between the active interactive vehicle and the target vehicle for,
[0032] ,
[0033] in, and These represent the longitudinal position coordinates of the active interactive vehicle and the target vehicle in the same longitudinal reference coordinate system, respectively. When the map information is reliable, the longitudinal position coordinates are the arc length coordinates projected onto the common reference route. When the map information is unreliable, the longitudinal position coordinates are the longitudinal coordinates projected onto the longitudinal unit direction of the local road.
[0034] when When the longitudinal positional relationship is determined, the active interactive vehicle cuts in front of the target vehicle; when When the longitudinal positional relationship is determined, the actively interacting vehicle merges into the target vehicle from behind; when When this is the case, the vertical positional relationship is determined as vertical overlap; where, This is the threshold for the longitudinal position difference.
[0035] Specifically, in step 2, the short-term motion path is determined by the vehicle. The predicted location trajectory within the preset prediction time range represents the vehicle's position trajectory. At the time of detection After the predicted time The predicted location is:
[0036] ,
[0037] in, Indicates vehicle At any moment The planar position, Indicates vehicle At any moment speed, Indicates the unit direction of travel determined by the vehicle's heading. Indicates the predicted time;
[0038] The method for selecting candidate interactive vehicle pairs is as follows: when the vehicle With vehicles If the short-term motion paths of both vehicles intersect within a preset prediction range, and the predicted arrival times of both vehicles at the intersection are positive, then the vehicles will be paired. Marked as candidate interactive vehicle pairs.
[0039] Specifically, in step 3, the candidate interactive vehicles... At any moment Interaction severity for,
[0040] ,
[0041] in, , For candidate interactive vehicles At any moment The acceleration is calculated by the following optimization problem:
[0042] Optimization Objective ,in, and The acceleration to be determined; and The predicted time for the two vehicles to reach the intersection point;
[0043] Under the assumption of constant acceleration, the two cars in time and The theoretical displacements within are as follows:
[0044] , ,in, and Representing vehicles and vehicles Theoretical displacement under the assumption of constant acceleration and Representing vehicles and vehicles Speed at the moment of detection;
[0045] The time difference between the arrival of the two vehicles at the intersection of their short-time motion paths satisfies ,in, To preset the minimum safe time threshold;
[0046] set up and vehicles and vehicles The error between the theoretical displacement of the two vehicles and the path distance from the current location to the predicted intersection point satisfies the following:
[0047] , ,in, The preset displacement error threshold is used;
[0048] When the two vehicles reach the intersection of their short-time motion paths, their vehicle speeds satisfy the following conditions: , ,and , ;
[0049] The method for determining the interaction time period is to continuously satisfy... The sampling times are merged into the same interaction time period, and the interaction time period is obtained after removing the time periods whose duration is not greater than the shortest duration threshold. ;
[0050] The method for determining the key interactive vehicle pair is as follows: when the interaction time period includes only two vehicles, the vehicle pair is directly designated as the key interactive vehicle pair; when the interaction time period involves three or more vehicles, the key interactive vehicle pair is determined as follows: Represented as,
[0051] ,
[0052] in, This represents the set of candidate vehicle pairs within the interaction time period; candidate vehicle pairs within the interaction time period Intra-interaction intensity .
[0053] Specifically, in step 4, the vehicle will be... The map information is determined to be reliable; otherwise, the vehicle... The map information was determined to be unreliable.
[0054] ,
[0055] Among them, vehicles Map matching success rate for,
[0056] ,in, This is an indicator function that takes the value 1 if the condition within the parentheses is true, and 0 otherwise. Indicates vehicles within the commonly observable time period The number of trajectory points, Indicates the first Each trajectory point can be matched with a valid lane; vehicle Median distance to the center line of the matched lane for ,in Indicates vehicle In the The distance from the center line of the matched lane at each trajectory moment. This represents median operations; This is the minimum map matching rate threshold. This represents the maximum allowed map matching distance. The table shows vehicles. The initial stable lane; It is an empty set;
[0057] The method for constructing the original route set, original driving corridor, and adjacent lane set for each vehicle is as follows:
[0058] With vehicles Starting from the initial stable lane, search forward along the lane successor connection relationship to obtain the original route set. The original driving corridor is obtained by adding a preset number of front-drive lanes to the original route set. The adjacent lane set is obtained based on the left-right adjacency relationship of each lane in the original route set. .
[0059] Specifically, in step 5, the strong map determination is performed using the following formula:
[0060] , ,
[0061] in, , vehicles and vehicles The result of the strong map determination is 1, which means that the vehicle meets the strong map determination conditions, and 0 means that it does not meet the conditions.
[0062] Indicates vehicle Is the vehicle able to enter stably? The original route, Indicates vehicle Is the vehicle able to enter stably? The original route;
[0063] Indicates to vehicles In terms of vehicles Whether it can stably maintain its original route; Indicates to vehicles In terms of vehicles Whether it can stably maintain its original route.
[0064] Specifically, vehicles Is the vehicle able to enter stably? The original route is determined by the following formula.
[0065] ,
[0066] in, Indicates vehicle Is the vehicle able to enter stably? The original route is used for role recognition on the observed trajectory. Indicates the determination of the vehicle The vehicle has been successfully entered. The original route; This indicates that the stable entry criteria have not been met; Indicates whether a vehicle has been detected. Continuous entry into vehicles Valid entry events for the original route; Indicates vehicle The final stable lane; Indicates vehicle The original set of routes; Indicates vehicle Enter the vehicle The number of consecutive trajectory frames maintained on the original route after the original route; This represents the minimum number of stable frames required to enter the opponent's route; This indicates the time from entry to the end of the observation period for the vehicle. Located in the vehicle The proportion of trajectory frames in the original route; This represents the minimum threshold value for maintaining the proportion of the route. For vehicles The criterion for determining whether a reliable lateral lane change has occurred is set to 1, indicating a reliable lateral lane change has occurred, and 0, indicating no reliable lateral lane change has occurred. This is calculated using the following formula:
[0067] ,
[0068] in, Indicates vehicle The map information reliability flag is set to 1 when the map information is reliable and 0 when the map information is unreliable. Indicates vehicle To ensure a stable and continuous matching to the same lane at the end of the trajectory, it is preferable to have at least 5 consecutive stable frames. Indicates vehicle The final stable lane; Indicates vehicle The original set of routes; Indicates vehicle The set of adjacent lanes corresponding to the original route; Indicates vehicle Whether there is evidence of continuous geometric lateral movement; when the vehicle has evidence of continuous geometric lateral movement. ,otherwise The actual determination considers both the lateral offset change and the consistency of the lateral movement direction; preferably, when the lateral offset change... And the lateral movement direction is consistent. hour, ,otherwise Specifically, this means that the net lateral change should be large enough, and at least 60% of the effective difference should be in the same direction as the overall lateral shift.
[0069] vehicle Whether it can stably maintain its original route is determined by the following formula.
[0070] ,
[0071] in, Indicates vehicle Whether the vehicle stably maintains its original route: 1 indicates that the vehicle stably maintains its original route, and 0 indicates that the condition for stably maintaining the original route is not met. Indicates vehicle The trajectory point is located in its original driving corridor. Other proportions; Indicates the maximum permitted percentage of departure from the original driving corridor; Indicates vehicle The number of times significant consecutive lateral transitions occurred; Indicates vehicle Continuously located in the original driving corridor The maximum number of trajectory frames other than; This represents the maximum number of frames allowed to continuously leave the original driving corridor.
[0072] An interactive vehicle role recognition system for autonomous driving scenarios includes,
[0073] The information acquisition module is configured to acquire vehicle trajectory information and road map information;
[0074] The candidate interactive vehicle pair filtering module is configured to traverse the moving vehicle pairs at the same time, predict short-term movement paths based on vehicle trajectory information, and filter out candidate interactive vehicle pairs.
[0075] The key interactive vehicle pair determination module is configured to calculate the interaction severity of candidate interactive vehicle pairs, divide the time when the interaction severity threshold is continuously met into an interaction time period, and determine the key interactive vehicle pairs within the interaction time period.
[0076] The map information processing module is configured to take the intersection of the timestamps of the key interactive vehicles as the common observable time period, perform time alignment and lane matching on the trajectories of the two key interactive vehicles, determine whether the map information of the vehicles is reliable, and construct the original route set, original driving corridor and adjacent lane set of each vehicle respectively.
[0077] The role recognition module is configured to perform strong map judgment based on the reliability of map information for each vehicle, combined with the original route set, original driving corridor and adjacent lane set of each vehicle, and identify active interactive vehicles and target vehicles based on the judgment results.
[0078] By adopting the above scheme, this invention filters key vehicle pairs based on the severity of the conflict, prioritizes explicit map evidence, uses map initiative scoring when map evidence is insufficient, and employs kinematic scoring in the local road coordinate system when map information is unavailable. Actively interacting vehicles and target vehicles are only output when the initiative differences meet reliability requirements; otherwise, they are marked as ambiguous scenarios. This invention enables the filtering of key interactive vehicles in complex cutting, lane changing, and merging scenarios, as well as the role recognition of actively interacting vehicles and target vehicles, improving the reliability and consistency of interactive scenario labeling and providing support for behavior prediction, risk assessment, and autonomous driving decision-making and control.
[0079] The present invention has the following technical effects:
[0080] First, the present invention characterizes the maneuver intensity required for a vehicle pair to reach a preset minimum safe time interval by using the minimum acceleration adjustment amount, and selects the vehicle pair with the highest maximum interaction severity as the key interaction vehicle in a multi-vehicle scenario, which can reduce the problem of selecting the wrong main interaction object by fixed distance threshold or nearest neighbor rule.
[0081] Second, this invention prioritizes the use of strong map evidence to identify roles, which is characterized by "one party making a credible lane change and stably entering the other party's route while the other party stably maintains its original route," thus giving the role determination clear road semantics and strong interpretability.
[0082] Third, this invention separates the lateral convergence phenomenon shared by two vehicles from the vehicle-level active contribution. It identifies the party that mainly causes the change in lateral relationship by features such as lateral approach contribution, deviation from its own route, and entry into the other party's route, thus avoiding the determination of the role based solely on the final position or front-back relationship.
[0083] Fourth, when map information is unreliable, this invention establishes a coordinate system based on local road directions, avoiding distortion of lateral movement of curved road sections caused by directly using global coordinate axes, and improving applicability under different road geometry conditions.
[0084] Fifth, this invention sets multiple role acceptance conditions and removes ambiguous samples, while separating the active role determination from the longitudinal sequential relationship, which can improve the accuracy of role labels in autonomous driving training data and reduce the impact of erroneous labels on behavior prediction and decision control models. Attached Figure Description
[0085] Figure 1 This is a flowchart of the method of the present invention;
[0086] Figure 2 This is a schematic diagram of the active interactive vehicle cutting in from the front of the target vehicle in an embodiment of the present invention;
[0087] Figure 3 This is a schematic diagram of an actively interacting vehicle merging into the target vehicle from behind, according to an embodiment of the present invention. Detailed Implementation
[0088] To further understand the present invention, preferred embodiments are described below with reference to the accompanying drawings and specific examples. It should be understood that the following embodiments are only for illustrating the technical solutions of the present invention and are not intended to limit the scope of protection of the present invention. Equivalent substitutions or adjustments made by those skilled in the art to the implementation steps, parameter thresholds, data sources, or feature combinations without departing from the concept of the present invention should all fall within the scope of protection of the present invention.
[0089] This invention addresses autonomous driving scenarios involving lane changes, ramp merging, and lane entry. It filters key interacting vehicle pairs from multi-vehicle trajectories and further identifies the active interacting vehicle initiating lateral relationship changes and its target vehicle. This embodiment uses a PM-type scenario from the publicly available autonomous driving natural trajectory dataset InterHub for illustration. The PM-type scenario includes major two-vehicle interactions formed during vehicle lane entry, merging, or lane changes. The specific method flow is as follows: Figure 1 As shown, it includes the following steps:
[0090] Step A: Obtain vehicle trajectory information and road map information; collect vehicle identification, timestamp, planar position, speed, and heading of each vehicle in the traffic scene, and sort the vehicle trajectories by time, handle outliers, and filter motion states; collect lane center lines, lane connection relationships, left and right adjacent relationships, and road driving directions;
[0091] Step B: Iterate through the pairs of moving vehicles at the same moment, predict their short-term movement paths based on the current position, speed and heading of each vehicle, and filter out candidate interactive vehicle pairs whose predicted paths have valid intersections;
[0092] Step C: For each candidate interactive vehicle pair, with the constraint that the time difference between the two vehicles reaching the intersection of the predicted paths is not less than a preset minimum safe time threshold, solve for the minimum acceleration adjustment required to satisfy the safe time interval. Calculate the interaction severity based on the sum of the absolute values of the acceleration adjustments of the two vehicles, and calculate the time difference between the predicted intersections of the two vehicles. Form a continuous interaction time period based on the interaction severity. When the scene contains multiple vehicles, select the vehicle pair with the highest maximum interaction severity within the continuous interaction time period as the key interactive vehicle pair.
[0093] Step D: Extract the common observable time period of key interactive vehicle pairs, perform time alignment and map matching on the trajectories of the two vehicles, and construct the initial routes, route corridors and adjacent lane sets of the two vehicles respectively.
[0094] Step E: Based on lane matching results, continuous lateral movement, reliable lane change, evidence of entering another vehicle's original route and maintaining its own original route, perform strong map determination, and determine whether the active interaction vehicle and the target vehicle can be uniquely identified according to the strong map determination rules; if and only if one key interaction vehicle stably enters the original route of another key interaction vehicle, and the other key interaction vehicle stably maintains its own original route, the former vehicle is identified as the active interaction vehicle, and the latter vehicle is identified as the target vehicle.
[0095] Step F: When the strong map determination rule cannot uniquely determine the role and the map information of at least one vehicle is reliable, calculate the map initiative score of the two vehicles respectively, and determine whether the role is reliable based on the lowest initiative score, absolute score difference, relative score ratio and lateral proximity contribution ratio.
[0096] Step G: When the map information of both vehicles is unreliable, establish a local road coordinate system based on the direction of the lane centerline, the direction of the centerline of the nearby road, or the direction of the vehicle's early trajectory fitting. Calculate the contribution of each vehicle to reducing the lateral distance between the two vehicles, the amount of lateral movement of the vehicles, and the change in heading to obtain the kinematic initiative score and determine whether the role is reliable.
[0097] Step H: When any decision level can uniquely identify the active party, the party that mainly causes the change in the lateral relationship between the two parties through its own lateral movement is identified as the active interactive vehicle, and the other key interactive vehicle is identified as the target vehicle; when both key interactive vehicles meet the strong map active condition, the map active score or the kinematic active score fails the role acceptance condition, or there is insufficient common observable trajectory, the scene is marked as an ambiguous scene, and the ambiguous scene is removed from the training samples that need to identify the active interactive vehicle, and the role is not forcibly assigned.
[0098] Step 1: After the role is determined, when the vehicle enters the other party's route or the lateral distance between the two parties reaches the interaction threshold, determine the longitudinal positional relationship between the active interactive vehicle and the target vehicle, and further output the task type, role source, confidence level and sample screening results.
[0099] Example 1: Interactive Vehicle Role Recognition Method
[0100] Step A: Obtain vehicle trajectory information and road map information
[0101] Acquire the trajectory information and road map information of each vehicle in the traffic scene to be processed. Vehicle trajectory information includes at least vehicle identification, timestamp, planar position, speed, and heading angle; road map information includes at least lane centerlines, lane connectivity, left and right adjacent lanes, and road direction of travel. For vehicles... At any given moment Position, speed, and direction of travel are respectively denoted as , and .
[0102] Step B: Filter candidate interactive vehicle pairs
[0103] Iterate through pairs of moving vehicles at the same moment, predicting their short-time motion paths based on each vehicle's current position, speed, and heading. The short-time motion paths are determined by the vehicles... The predicted location trajectory within the preset prediction time range represents the vehicle's position trajectory. At the time of detection After the predicted time The position is represented as:
[0104]
[0105] in, Indicates vehicle The predicted location, This indicates the prediction time. If the predicted paths of two vehicles intersect within a preset prediction range, and the arrival times of both vehicles at the intersection are positive, then the vehicle pair is identified as a candidate interactive vehicle pair.
[0106] Step C: Calculate the severity of the interaction and identify key interaction vehicle pairs.
[0107] For candidate vehicles With the objective of minimizing the sum of the absolute values of acceleration required for the time difference between the two vehicles reaching the intersection of the predicted paths to be no less than a preset minimum safe time threshold, the following optimization problem is constructed:
[0108]
[0109] in, and Representing vehicles and vehicles The acceleration to be determined and Representing vehicles and vehicles The time to reach the predicted intersection point.
[0110] Under the assumption of constant acceleration, the vehicle and vehicles In time and The theoretical displacements within are as follows:
[0111] ,
[0112] in, and Representing vehicles and vehicles Theoretical displacement under the assumption of constant acceleration and Representing vehicles and vehicles The speed at the moment of detection.
[0113] And satisfy:
[0114]
[0115] ,
[0116] , ,
[0117] in, and Representing vehicles and vehicles The path distance from the current location to the predicted intersection point To preset the minimum safe time threshold, A preset displacement error threshold is set. Preferably, .
[0118] Let the optimized accelerations be respectively and Then the vehicle is at time The severity of the interaction is defined as follows:
[0119]
[0120] The time difference between the arrival of the two vehicles at the predicted intersection point is defined as:
[0121]
[0122] in, and They represent the times at time 1 and 2 respectively. The vehicle obtained after optimization and vehicles The time to reach the predicted path intersection point; the greater the interaction severity, the stronger the acceleration or deceleration adjustment required to make the time difference between the two vehicles reaching the predicted path intersection point reach the minimum safe time interval, i.e., the more urgent the potential interaction between the vehicle pair. The predicted intersection point arrival time difference is used to record the time interval between the two vehicles reaching the predicted path intersection point after optimization; the urgency of the interaction between the vehicle pair is mainly characterized by the interaction severity required to reach the minimum safe time interval.
[0123] Will satisfy The consecutive moments are divided into the same interactive time period. Preferably, ; Remove items whose duration is not greater than the shortest duration threshold The time period is used to obtain the interaction time period. Preferably, .
[0124] Vehicles during the interaction period The maximum interaction strength within is defined as:
[0125]
[0126] When the interaction period contains only two vehicles, that vehicle pair is directly identified as the key interaction vehicle pair; when the interaction period contains three or more vehicles, the candidate vehicle pair set is used. Select the vehicle pair with the highest interaction intensity:
[0127]
[0128] The remaining vehicles are used as auxiliary vehicles for recording.
[0129] Step D: Align common trajectories and determine the reliability of map information
[0130] Let the two key interactive vehicles obtained from the screening be vehicles. and vehicles The intersection of the timestamps of the two vehicles is used as the common observable time period, and the position, speed, and heading information of the two vehicles are extracted based on the same timestamp. If the number of common observable frames or the common observation duration is less than a preset threshold, the scene is marked as insufficient evidence.
[0131] For vehicles Lane matching is performed on each trajectory point, and the map matching success rate is:
[0132]
[0133] The median distance from the lane centerline is:
[0134]
[0135] When the following conditions are met simultaneously, the vehicle will be... The map information is marked as reliable:
[0136]
[0137] in, Indicates vehicles within the commonly observable time period The number of trajectory points, Indicates the first Whether each trajectory point has been successfully matched with a valid lane. Indicates vehicle The initial stable lane. Preferably, , .
[0138] With vehicles Starting from the initial stable lane, the search proceeds forward along the lane successor connections to obtain the original route set. Adding a preset number of front-drive lanes to the original route set yields the original driving corridor. The set of adjacent lanes is obtained based on the left-right adjacency relationship. By including normal subsequent lanes within the original route set, changes in lane segment markings within the same route are avoided from being misinterpreted as lateral lane changes.
[0139] Step E: Determine the role based on strong map evidence
[0140] The vehicle trajectory is projected onto its original route centerline to obtain the signed lateral offset of the vehicle at each trajectory time. Directional statistics are performed on the lateral offset differences between adjacent trajectory points to obtain the consistency of the vehicle's lateral movement direction. The lateral offset change and the consistency of the lateral movement direction are as follows:
[0141]
[0142]
[0143] in, This represents the number of differences that indicate the lateral offset change reaches the effective change threshold. This indicates the number of effective differences whose direction is consistent with the overall lateral movement direction. Preferably, when... At that time, it was believed that the vehicle had evidence of continuous geometric lateral motion.
[0144] vehicle Trusted lane change sign At least the following conditions must be met simultaneously: the map information is reliable; the number of consecutive frames that stably match the same lane in the final segment is not less than a preset value of 5 frames; and the stable lane in the final segment does not belong to the original route set. The final stable lane belongs to the set of adjacent lanes. Evidence of continuous geometric lateral motion of the vehicle.
[0145] vehicle Stable entry into the vehicle The original route is marked as ,in, Indicates the determination of the vehicle The vehicle has been successfully entered. The original route; This indicates that the stable entry criteria have not been met. At least the following conditions must be met simultaneously: vehicle... A reliable lane change has occurred; the vehicle First consecutive entry into the vehicle The number of frames for the original route is not less than the preset value; vehicles The final stable lane belongs to vehicles The original route; after entering, maintain a frame rate no less than the preset value; after entering, be located in the vehicle. The proportion of trajectory points within the original route is not less than a preset proportion. Preferably, after a vehicle enters another vehicle's route, it maintains its position for at least 5 consecutive frames, and from the entry time to the end of the observation, at least 80% of the trajectory frames are located within the original route of the other vehicle. The indicator of a vehicle stably maintaining its original route is denoted as... It must at least satisfy the following conditions: the final stable lane belongs to the original route set; it leaves the original driving corridor. The trajectory proportion does not exceed 15%; there are no continuous significant lateral lane changes; the number of frames continuously leaving the original driving corridor is less than 5 frames.
[0146] vehicle Maintain stable course markers At least the following conditions must be met simultaneously: the vehicle map information is reliable; the final stable lane belongs to its original route; the proportion of the trajectory deviating from its own route corridor is not greater than a preset proportion; there is no continuous significant lateral lane change; and the longest number of frames continuously deviating from its own route corridor is less than a preset value.
[0147]
[0148] in, Indicates vehicle Whether the vehicle stably maintains its original route: 1 indicates that the vehicle stably maintains its original route, and 0 indicates that the condition for stably maintaining the original route is not met. Indicates vehicle The trajectory point is located in its original driving corridor. Other proportions; Indicates the maximum permitted percentage of departure from the original driving corridor; Indicates vehicle The number of times significant consecutive lateral transitions occurred; Indicates vehicle The maximum number of consecutive trajectory frames located outside the original driving corridor; This represents the maximum number of frames allowed to continuously leave the original driving corridor. Preferably, That is, the proportion of the vehicle's trajectory outside the original driving corridor does not exceed 15%, and the number of consecutive frames outside the original driving corridor is less than 5 frames.
[0149] For the two key interactive vehicles and The active conditions for a strong map are defined as follows:
[0150]
[0151] In the formula, , vehicles and vehicles The result of the strong map determination is 1, which means the vehicle meets the strong map determination conditions, and 0 means it does not. and At that time, the vehicle Identify it as an actively interacting vehicle and move the vehicle Identified as the target vehicle; when and At times, the roles are reversed; when When, mark the scene as an ambiguous scene; when At that time, the initiative scoring stage begins.
[0152] Step F: Map Initiative Scoring
[0153] When map information for at least one key interactive vehicle is reliable, but strong map conditions cannot uniquely determine the role, calculate the vehicle's map initiative score. Map initiative score is composed of reliable lane changes, stable entry into the opponent's route, lateral approach contribution, deviation from one's own route, and heading changes:
[0154]
[0155] in, This indicates the vehicle's contribution to the reduction in lateral spacing. This indicates the lateral deviation of the vehicle from its original path. This indicates the change in heading angle. Preferably: , The weight of discrete map events is greater than the weight of continuous motion features to avoid the same lane-changing behavior being repeatedly amplified by multiple related continuous features.
[0156] Step G: Kinematic initiative scoring
[0157] When both vehicle map information is unreliable, a local road coordinate system is established based on the fitted direction of the vehicles' early trajectories or the direction of the centerline of nearby roads. The lateral unit vector is obtained by rotating the vertical unit vector by 90 degrees. The local vertical and lateral unit vectors are as follows:
[0158] ,
[0159] The projection of the vehicle's position in the local lateral direction is:
[0160]
[0161] Let the initial lateral relative positions of the two vehicles be:
[0162]
[0163] Then the vehicle and vehicles The lateral approach contributions towards the other are as follows:
[0164]
[0165]
[0166] vehicle The kinematic initiative score is:
[0167]
[0168] in, This represents the net lateral displacement of the vehicle. Preferably, , .
[0169] The kinematic score uses the same minimum score, absolute difference, relative proportion, and lateral proximity contribution percentage as the map initiative score to determine role acceptance.
[0170] Step H: Determine role differences and filter ambiguous samples.
[0171] Set up vehicles The vehicle's initiative score and the vehicle's initiative score are respectively and ,but:
[0172] ,
[0173] ,
[0174] in, This indicates the ratio between higher initiative scores and lower initiative scores. The larger the value, the more significant the difference in initiative between the two vehicles.
[0175] The percentage of lateral approach contribution for vehicles with higher scores is:
[0176]
[0177] Preferably, The threshold can be calibrated based on manually labeled samples, road type, and trajectory sampling frequency. Vehicles with higher scores are identified as actively interacting vehicles only when the following conditions are met simultaneously:
[0178]
[0179] Otherwise, the scenario is marked as ambiguous, the active interactive vehicle identifier and the target vehicle identifier are set to empty, and the scenario is removed from the active interactive vehicle training sample. For scenarios where the active interactive vehicle can be reliably identified, another key interactive vehicle is identified as the target vehicle.
[0180] Step 1: Determine the vertical positional relationship and output the result.
[0181] In strong map-based scenarios, the interaction moment is defined as the instant when the actively interacting vehicle stably enters the target vehicle's original route; in map-based proactive scoring or kinematic regression scenarios, the interaction moment is defined as the moment when the lateral distance between the two vehicles first becomes less than the target vehicle's original path. The moment when the two vehicles are at their minimum lateral distance is used as the interaction moment. If such a moment does not exist, the moment when the lateral distance between the two vehicles is at its minimum is used.
[0182] At the moment of interaction determination, the positions of the two vehicles are projected onto a common reference route or a local longitudinal direction, and the longitudinal position difference between the actively interacting vehicle and the target vehicle is calculated:
[0183]
[0184] when hour, This represents the longitudinal position difference threshold used to eliminate positioning errors and slight longitudinal fluctuations, determining when the actively interacting vehicle cuts in front of the target vehicle; when When the vehicle actively interacting is merging behind the target vehicle, it is determined that the vehicle is merging; otherwise, it is determined to be longitudinally overlapping. Preferably, The described longitudinal positional relationship is only used to describe the interaction mode and does not participate in the selection of the actively interacting vehicle.
[0185] when When, it is determined that the actively interacting vehicle cuts in front of the target vehicle; when When the vehicle actively interacting merges into the target vehicle from behind, it is determined that the vehicle is merging into the target vehicle. When the two vehicles are in a longitudinally overlapping state, the final output includes the active interaction vehicle number, the target vehicle number, the source of the role determination, the confidence level, the longitudinal positional relationship, the ambiguity marker, and the sample selection results.
[0186] As a further optimization of the interactive vehicle role recognition method for autonomous driving entry scenarios described in this invention, the task type and role determination are independent of each other. When the role is reliably determined and an active interactive vehicle enters the scene, the task type is marked as active entry; when another vehicle is confirmed to have active entry behavior and the currently controlled vehicle is the interacting party, the task type is marked as passive entry; when both vehicles have positive evidence of stably maintaining their own routes and there is no obvious lateral convergence, the task type is marked as following; other scenarios with insufficient evidence are marked as unresolved interaction.
[0187] Example 2: Frontal approach scenario with strong map evidence
[0188] Figure 2The vehicle trajectories and road maps for scene-84632 are shown. Through steps B and C, candidate interaction detection and interaction severity ranking were performed on vehicle pairs within the scene, identifying vehicles 7 and 11 as the key interactive vehicle pair. During the jointly observable time period for both vehicles, the map matching results for vehicles 7 and 11 both met the map reliability conditions.
[0189] The original route sets and original travel corridors were constructed using the initial stable lanes of vehicles 7 and 11, respectively. Vehicle 7's trajectory shows evidence of continuous geometric lateral movement; its final stable lane does not belong to vehicle 7's original route, but rather to an adjacent lane of vehicle 7's original route. Simultaneously, vehicle 7 stably enters vehicle 11's original route. Vehicle 11 maintains its original route throughout the shared observation period without significant continuous lateral transitions. Therefore:
[0190]
[0191]
[0192] Vehicle 11 does not meet the conditions for stably entering the original route of vehicle 7, therefore Based on this, vehicle 7 was identified as the active interactive vehicle and vehicle 11 as the target vehicle, using strong map evidence. The role determination was based on strong map evidence with a high confidence level. Furthermore, by comparing the longitudinal positions of the two vehicles at the moment vehicle 7 entered vehicle 11's original route, it was found that vehicle 7 was in front of vehicle 11, thus determining the longitudinal positional relationship as a forward cut-in. The roles in this scenario are clear, and it is retained as a valid sample.
[0193] Example 3: Rear-entry scenario under insufficient map information
[0194] Figure 3 The vehicle trajectories and road map for scene-63041 are shown. Through candidate interaction detection and interaction severity ranking, vehicles 455 and 476 were identified as the key interaction vehicle pair. Due to the complex local road topology in this scene, the map matching rate of the vehicle trajectories, lane centerline distance, or initial stable lane conditions failed to simultaneously meet the map reliability requirements. Therefore, strong map evidence for credible lane changing, entry into the target route, and the target vehicle maintaining its route could not be formed, thus triggering a kinematic regression decision.
[0195] A local road coordinate system is established based on the early motion directions of the two vehicles. After projecting the trajectories of the two vehicles onto the local lateral direction, the lateral approach contribution and kinematic initiative score of vehicle 455 and vehicle 476 are calculated respectively. The difference in initiative scores between the two vehicles is output by the program:
[0196]
[0197] The contribution percentage of horizontal proximity is:
[0198]
[0199] Therefore, we can conclude that:
[0200]
[0201] Based on the highest initiative score and score ratio output by the program, all the acceptance conditions for the roles are met. Therefore, vehicle 455 is identified as the active interaction vehicle, and vehicle 476 is identified as the target vehicle. Since this result comes from kinematic regression under conditions of insufficient map information, the role confidence is recorded as low. The moment when the lateral distance between the two vehicles first falls below a preset threshold is taken as the interaction moment. Comparing the positions of the two vehicles along the local longitudinal direction, it is found that vehicle 455 is behind vehicle 476, thus the longitudinal positional relationship is determined to be rear-merging. Although this scenario lacks reliable map evidence, the kinematic initiative differences are clear, therefore it is retained as a valid sample.
[0202] Recognition results for two typical scenarios:
[0203] Table 1 provides the following information: Figure 2 and Figure 3 The results of interactive vehicle role recognition are shown in two typical entry scenarios. Figure 2 The scene uses strong map evidence to determine the characters; Figure 3 When map information is insufficient, the scene completes the role determination through kinematic backtracking, indicating that the present invention can identify interactive vehicle roles by adopting corresponding evidence paths for different map information qualities.
[0204] Table 1
[0205]
[0206] This invention also provides an interactive vehicle role recognition system for autonomous driving entry scenarios, used to implement the aforementioned interactive vehicle role recognition method for autonomous driving entry scenarios, including,
[0207] The information acquisition module is configured to acquire vehicle trajectory information and road map information;
[0208] The candidate interactive vehicle pair filtering module is configured to traverse the moving vehicle pairs at the same time, predict short-term movement paths based on vehicle trajectory information, and filter out candidate interactive vehicle pairs.
[0209] The key interactive vehicle pair determination module is configured to calculate the interaction severity of candidate interactive vehicle pairs, divide the time when the interaction severity threshold is continuously met into an interaction time period, and determine the key interactive vehicle pairs within the interaction time period.
[0210] The map information processing module is configured to take the intersection of the timestamps of the key interactive vehicles as the common observable time period, perform time alignment and lane matching on the trajectories of the two key interactive vehicles, determine whether the map information of the vehicles is reliable, and construct the original route set, original driving corridor and adjacent lane set of each vehicle respectively.
[0211] The role recognition module is configured to perform strong map judgment based on the reliability of map information for each vehicle, combined with the original route set, original driving corridor and adjacent lane set of each vehicle, and identify active interactive vehicles and target vehicles based on the judgment results.
[0212] In summary, this invention is used to identify the active interactive vehicle and the target vehicle during the interaction process, providing support for behavior prediction, risk assessment, and decision control of autonomous driving systems.
[0213] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.
[0214] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.
Claims
1. A method for interactive vehicle role recognition in autonomous driving scenarios, characterized in that... Includes the following steps: Step 1: Obtain vehicle trajectory information and road map information; Step 2: Iterate through pairs of moving vehicles at the same time, predict short-term movement paths based on vehicle trajectory information, and filter to obtain candidate interactive vehicle pairs; Step 3: Calculate the interaction severity of candidate interactive vehicle pairs, divide the moments that continuously meet the interaction severity threshold into interaction time periods, and identify key interactive vehicle pairs within the interaction time periods. Step 4: Take the intersection of the timestamps of the key interactive vehicles as the common observable time period, perform time alignment and lane matching on the trajectories of the two key interactive vehicles in the common observable time period, determine whether the vehicle map information is reliable, and construct the original route set, original driving corridor and adjacent lane set of each vehicle respectively. Step 5: Based on the judgment results of whether the map information of each vehicle is reliable, and combined with the original route set, original driving corridor and adjacent lane set of each vehicle, a strong map judgment is made, and the active interactive vehicle and target vehicle are identified based on the judgment results.
2. The method as described in claim 1, characterized in that: In step 5, the active interactive vehicle and the target vehicle are identified based on the following conditions of the determination result. Case 1: If and only if one vehicle meets the strong map determination criteria, the vehicle that meets the strong map determination criteria will be identified as the active interaction vehicle, and the other vehicle will be identified as the target vehicle. Case 2: If neither vehicle meets the criteria for a strong map, and the map information of at least one key interactive vehicle is reliable, the active interactive vehicle and the target vehicle are determined based on the map initiative scores of the two vehicles. Case 3: If neither vehicle meets the strong map determination criteria and the map information of both key interactive vehicles is unreliable, the active interactive vehicle and the target vehicle are determined based on the kinematic initiative scores of the two vehicles. Case 4: When both vehicles meet the strong map determination criteria, the scene is marked as an ambiguous scene, and the ambiguous scene is removed from the training samples that need to determine the active interaction vehicle.
3. The method as described in claim 2, characterized in that: The method for determining the active interaction vehicle and the target vehicle based on the map initiative score in scenario 2 or the kinematic initiative score in scenario 3 is as follows: The vehicle with the higher map initiative score or kinematic initiative score is identified as the active interaction vehicle, and the other vehicle is identified as the target vehicle, provided that the following conditions are met simultaneously: , , , , in, The larger of the two vehicle initiative scores is taken as the initiative score, which can be either the map initiative score or the kinematic initiative score. The smaller of the two vehicle initiative scores is taken as the initiative score, which can be either the map initiative score or the kinematic initiative score. This indicates the minimum initiative score required to confirm the active interaction of vehicles. This represents the minimum absolute difference that the initiative scores of the two vehicles must satisfy. This represents the minimum threshold for the ratio of the two vehicles' initiative scores. This indicates the minimum lateral contribution percentage required for vehicles to achieve a higher initiative score. and Representing vehicles and vehicles Contribution to reducing the lateral distance between the two vehicles; This indicates the lateral approach contribution of vehicles with higher initiative scores, when the vehicle When the initiative score is high, When the vehicle When the initiative score is high, , This indicates a preset positive number to prevent the denominator from being zero.
4. The method as described in claim 2 or 3, characterized in that: A scenario will be marked as ambiguous if any of the following conditions are met: both key interactive vehicles meet the strong map determination criteria; or, during the map initiative scoring or kinematic initiative scoring phase, the role acceptance criteria of minimum initiative score, absolute difference in initiative scores, initiative score ratio, and lateral proximity contribution ratio are not met simultaneously; or, the common observable trajectories of the two key interactive vehicles are insufficient to support role determination. When a scene is marked as ambiguous, the active interactive vehicle identifier and the target vehicle identifier are set to empty, the ambiguity flag is set to true, and the ambiguous scene is removed from the training samples that need to determine the active interactive vehicle.
5. The method as described in claim 2, characterized in that: For scenarios where the active interactive vehicle and the target vehicle have been determined in scenarios 1, 2 and 3, the method further includes determining the interaction determination time and determining the longitudinal positional relationship between the active interactive vehicle and the target vehicle based on the interaction determination time. In Case 1, the moment when the active interactive vehicle enters the original route of the target vehicle is determined as the interaction determination moment; in Cases 2 and 3, the moment when the lateral distance between the active interactive vehicle and the target vehicle first falls below a preset threshold is determined as the interaction determination moment.
6. The method as described in claim 5, characterized in that: Based on the determination of the interaction timing, the longitudinal positional relationship between the actively interacting vehicle and the target vehicle is determined, including: If the interaction determination time has been determined, then the following calculations are performed using the interaction determination time; If the interaction determination time is not determined, the time when the lateral distance between the two vehicles is the smallest shall be used as the interaction determination time for the following calculation; At the moment of interaction determination, the longitudinal position difference between the active interactive vehicle and the target vehicle for, , in, and These represent the longitudinal position coordinates of the active interactive vehicle and the target vehicle in the same longitudinal reference coordinate system, respectively. When the map information is reliable, the longitudinal position coordinates are the arc length coordinates projected onto the common reference route. When the map information is unreliable, the longitudinal position coordinates are the longitudinal coordinates projected onto the longitudinal unit direction of the local road. when When the longitudinal positional relationship is determined, the active interactive vehicle cuts in front of the target vehicle; when When the longitudinal positional relationship is determined, the actively interacting vehicle merges into the target vehicle from behind; when When this is the case, the vertical positional relationship is determined as vertical overlap; where, This is the threshold for the longitudinal position difference.
7. The method as described in claim 1, characterized in that: In step 2, the short-term motion path is formed by the vehicle. The predicted location trajectory within the preset prediction time range represents the vehicle's position trajectory. At the time of detection After the predicted time The predicted location is: , in, Indicates vehicle At any moment The planar position, Indicates vehicle At any moment speed, Indicates the unit direction of travel determined by the vehicle's heading. Indicates the predicted time; The method for selecting candidate interactive vehicle pairs is as follows: when the vehicle With vehicles If the short-term motion paths of both vehicles intersect within a preset prediction range, and the predicted arrival times of both vehicles at the intersection are positive, then the vehicles will be paired. Marked as candidate interactive vehicle pairs.
8. The method as described in claim 1, characterized in that: In step 3, the candidate interactive vehicle pairs At any moment Interaction severity for, , in, , For candidate interactive vehicles At any moment The acceleration is calculated by the following optimization problem: Optimization Objective ,in, and The acceleration to be determined; and The predicted time for the two vehicles to reach the intersection point; Under the assumption of constant acceleration, the two cars in time and The theoretical displacements within are as follows: , , in, and Representing vehicles and vehicles Theoretical displacement under the assumption of constant acceleration and Representing vehicles and vehicles Speed at the moment of detection; The time difference between the arrival of the two vehicles at the intersection of their short-time motion paths satisfies ,in, To preset the minimum safe time threshold; set up and vehicles and vehicles The error between the theoretical displacement of the two vehicles and the path distance from the current location to the predicted intersection point satisfies the following: , ,in, The preset displacement error threshold is used; When the two vehicles reach the intersection of their short-time motion paths, their vehicle speeds satisfy the following conditions: , ,and , ; The method for determining the interaction time period is to continuously satisfy... The sampling times are merged into the same interaction time period, and the interaction time period is obtained after removing the time periods whose duration is not greater than the shortest duration threshold. ; The method for determining the key interactive vehicle pair is as follows: when the interaction time period includes only two vehicles, the vehicle pair is directly designated as the key interactive vehicle pair; when the interaction time period involves three or more vehicles, the key interactive vehicle pair is determined as follows: Represented as, , in, This represents the set of candidate vehicle pairs within the interaction time period; candidate vehicle pairs within the interaction time period Intra-interaction intensity .
9. The method as described in claim 1, characterized in that: In step 4, the vehicle will be... The map information is determined to be reliable; otherwise, the vehicle... The map information was determined to be unreliable. , Among them, vehicles Map matching success rate for, ,in, This is an indicator function that takes the value 1 if the condition within the parentheses is true, and 0 otherwise. Indicates vehicles within the commonly observable time period The number of trajectory points, Indicates the first Each trajectory point can be matched with a valid lane; vehicle Median distance to the center line of the matched lane for ,in Indicates vehicle In the The distance from the center line of the matched lane at each trajectory moment. This represents median operations; This is the minimum map matching rate threshold. This represents the maximum allowed map matching distance. The table shows vehicles. The initial stable lane; It is an empty set; The method for constructing the original route set, original driving corridor, and adjacent lane set for each vehicle is as follows: With vehicles Starting from the initial stable lane, search forward along the lane successor connection relationship to obtain the original route set. The original driving corridor is obtained by adding a preset number of front-drive lanes to the original route set. The adjacent lane set is obtained based on the left-right adjacency relationship of each lane in the original route set. .
10. The method as described in claim 1, characterized in that: In step 5, the strong map determination is performed using the following formula: , , in, , vehicles and vehicles The result of the strong map determination is 1, which means that the vehicle meets the strong map determination conditions, and 0 means that it does not meet the conditions. Indicates vehicle Is the vehicle able to enter stably? The original route, Indicates vehicle Is the vehicle able to enter stably? The original route; Indicates to vehicles In terms of vehicles Whether it can stably maintain its original route; Indicates to vehicles In terms of vehicles Whether it can stably maintain its original route.
11. The method as described in claim 10, characterized in that: vehicle Is the vehicle able to enter stably? The original route is determined by the following formula. , in, Indicates vehicle Is the vehicle able to enter stably? The original route, Indicates the determination of the vehicle The vehicle has been successfully entered. The original route; This indicates that the stable entry criteria have not been met; Indicates whether a vehicle has been detected. Continuous entry into vehicles Valid entry events for the original route; Indicates vehicle The final stable lane; Indicates vehicle The original set of routes; Indicates vehicle Enter the vehicle The number of consecutive trajectory frames maintained on the original route after the original route; This represents the minimum number of stable frames required to enter the opponent's route; This indicates the time from entry to the end of the observation period for the vehicle. Located in the vehicle The proportion of trajectory frames in the original route; This represents the minimum threshold value for maintaining the proportion of the route. For vehicles The criterion for determining whether a reliable lateral lane change has occurred is set to 1, indicating a reliable lateral lane change has occurred, and 0, indicating that the reliable lateral lane change criteria have not been met. This is calculated using the following formula: , in, Indicates vehicle The map information reliability flag is set to 1 when the map information is reliable and 0 when the map information is unreliable. Indicates vehicle Does the trajectory consistently and stably match the same lane at the end of the trajectory phase? Indicates vehicle The final stable lane; Indicates vehicle The original set of routes; Indicates vehicle The set of adjacent lanes corresponding to the original route; Indicates vehicle Whether there is evidence of continuous geometric lateral movement; when the vehicle has evidence of continuous geometric lateral movement. ,otherwise ; vehicle Whether it can stably maintain its original route is determined by the following formula. , in, Indicates vehicle Whether the vehicle stably maintains its original route: 1 indicates that the vehicle stably maintains its original route, and 0 indicates that the condition for stably maintaining the original route is not met. Indicates vehicle The trajectory point is located in its original driving corridor. Other proportions; Indicates the maximum permitted percentage of departure from the original driving corridor; Indicates vehicle The number of times significant consecutive lateral transitions occurred; Indicates vehicle Continuously located in the original driving corridor The maximum number of trajectory frames other than; This represents the maximum number of frames allowed to continuously leave the original driving corridor.
12. An interactive vehicle role recognition system for autonomous driving scenarios, characterized in that: include, The information acquisition module is configured to acquire vehicle trajectory information and road map information; The candidate interactive vehicle pair filtering module is configured to traverse the moving vehicle pairs at the same time, predict short-term movement paths based on vehicle trajectory information, and filter out candidate interactive vehicle pairs. The key interactive vehicle pair determination module is configured to calculate the interaction severity of candidate interactive vehicle pairs, divide the time when the interaction severity threshold is continuously met into an interaction time period, and determine the key interactive vehicle pairs within the interaction time period. The map information processing module is configured to take the intersection of the timestamps of the key interactive vehicles as the common observable time period, perform time alignment and lane matching on the trajectories of the two key interactive vehicles, determine whether the map information of the vehicles is reliable, and construct the original route set, original driving corridor and adjacent lane set of each vehicle respectively. The role recognition module is configured to perform strong map judgment based on the reliability of map information for each vehicle, combined with the original route set, original driving corridor and adjacent lane set of each vehicle, and identify active interactive vehicles and target vehicles based on the judgment results.
