Roadside sensing data quality real-time evaluation method based on high-precision map

Abstract

The embodiment of the invention discloses a roadside sensing data quality real-time evaluation method based on a high-precision map, and the method comprises the steps: obtaining related data of a predetermined region, carrying out the preprocessing, and carrying out the feature extraction of the preprocessed data; performing quality evaluation on the trajectory data by using the feature extraction result to obtain an evaluation result; and comprehensively analyzing the evaluation result to obtain a quality analysis report, and feeding back a comprehensive analysis result to the roadside sensing system. According to the method, the perception data is compared with the dynamic data in the high-precision map from multiple dimensions, so that the accuracy and practicability of the evaluation result are improved; in addition, a real-time feedback mechanism is established, problems in the trajectory data are found and corrected in time, and continuous optimization of the roadside sensing system is promoted. Therefore, the quality of the trajectory data in the roadside sensing system can be comprehensively and accurately evaluated, and continuous optimization of the system is promoted through a real-time feedback mechanism.

Description

Technical Field

[0001] The present application relates to the technical field of traffic data testing, and in particular to a real-time evaluation method for roadside perception data quality based on a high-precision map. Background Art

[0002] With the development of intelligent connected vehicles and autonomous driving technology, roadside perception systems play an important role in improving traffic safety and efficiency. These systems monitor the dynamic information of traffic participants in real time through sensors deployed on the side of the road, providing environmental perception support for vehicles. As the basis of autonomous driving, high-precision maps (HD Maps) contain detailed data such as road geometry, lane information, and traffic signs, providing vehicles with accurate positioning and navigation references.

[0003] However, the quality of roadside perception data directly affects the safety and reliability of the autonomous driving system. Existing quality assessment methods mainly focus on position accuracy and lack comprehensive assessment of multiple dimensions such as speed, path, and behavior, resulting in one-sided assessment results. In addition, traditional assessment methods mostly rely on simple error calculations and lack in-depth analysis of the quality of perception data in complex scenarios. At the same time, existing technologies have failed to establish a real-time feedback mechanism, and are unable to promptly detect and correct problems in perception data, affecting the system's self-optimization capabilities. Existing methods mainly use the static information of high-precision maps, and do not fully consider their dynamic characteristics, such as traffic lights, construction areas, etc., resulting in incomplete evaluation results.

[0004] The paper document 1 of the existing technical solution (Du Yuchuan, Du Zhouyang, Shi Yupeng, Zhao Cong, Ji Yuxiong. Intelligent evaluation method for roadside perception vehicle trajectory data quality [J]. China Journal of Highway and Transport, 2021, 34(7): 164-176) records an intelligent evaluation method for roadside perception vehicle trajectory data quality. This method is based on trajectory multivariate information, designs trajectory rationality, volatility and interaction abnormality evaluation indicators from three dimensions: element characteristics, temporal characteristics and spatial characteristics, and analyzes the correlation between the evaluation indicators and the trajectory data quality level. On this basis, an intelligent evaluation method is proposed to realize the intelligent evaluation of the quality of roadside perception vehicle trajectory data by constructing an evaluation model.

[0005] Patent document 2 (CN114357019B) of the prior art describes a method for monitoring the data quality of roadside perception units in an intelligent connected environment. This method uses intelligent connected vehicle data with higher perception accuracy to evaluate and monitor the data of roadside perception units. By matching the trajectory data corresponding to the intelligent connected vehicle in the roadside perception trajectory data set, aligning the system time of the roadside perception unit with the system time of the intelligent connected vehicle, and achieving time synchronization through trajectory resampling, the data quality of the roadside perception unit can be comprehensively, real-time and reliably monitored.

[0006] Patent document 3 (CN113642845B) of the prior art describes a method for quality assessment of road traffic perception trajectory data, which performs quality assessment on a non-interactive road traffic perception trajectory data set, uses a model to perform the assessment, and outputs the model quality assessment score as the assessment result.

[0007] These existing technical solutions mainly focus on position accuracy and static information, lack comprehensive evaluation of multiple dimensions such as speed, path, and behavior, and have not established a real-time feedback mechanism, making it impossible to comprehensively and accurately evaluate the quality of roadside perception data. Summary of the invention

[0008] This application provides a real-time evaluation method for roadside perception data quality based on a high-precision map to perform a real-time and comprehensive evaluation of the quality of trajectory data of a roadside perception system. The specific scheme is as follows:

[0009] A real-time evaluation method for roadside perception data quality based on a high-precision map includes:

[0010] Acquire relevant data of a predetermined area, wherein the relevant data includes dynamic data of a high-precision map and trajectory data of traffic participants in the predetermined area acquired from a roadside perception system;

[0011] Performing data preprocessing on the relevant data to obtain preprocessed data;

[0012] Performing feature extraction on the preprocessed data to obtain a feature extraction result;

[0013] Using the feature extraction result to perform quality assessment on the trajectory data to obtain an assessment result;

[0014] The evaluation results are comprehensively analyzed to obtain a comprehensive analysis result, and the comprehensive analysis result is fed back to the roadside perception system.

[0015] Optionally, the performing data preprocessing on the relevant data to obtain preprocessed data includes:

[0016] Performing data cleaning on the relevant data to remove missing, abnormal or erroneous data points to obtain cleaned data;

[0017] Synchronizing the timestamps of the cleaned data to obtain synchronized data;

[0018] The coordinate system of the synchronized data is converted to obtain data in a unified coordinate system, and the data is used as preprocessed data.

[0019] Optionally, the feature extraction results include: position features, speed features, path features and behavior features; the position features represent the position coordinates of traffic participants at different time points, the speed features represent the instantaneous speed and average speed of traffic participants, the path features represent the driving paths of traffic participants, and the behavior features represent the driving behaviors of traffic participants.

[0020] Optionally, the using the feature extraction result to perform quality assessment on the trajectory data to obtain an assessment result includes:

[0021] The feature extraction results are used to calculate the position accuracy score, path rationality score and behavior consistency score respectively, and the three are used as the evaluation results of the trajectory data.

[0022] Optionally, the using the feature extraction results to respectively calculate the position accuracy score, the path rationality score and the behavior consistency score, and using the three as the evaluation results of the trajectory data includes:

[0023] Evaluating the position accuracy by using the Euclidean distance between the first position coordinate represented by the position feature and the corresponding second position coordinate in the high-precision map;

[0024] Use the dynamic time warping algorithm to calculate the similarity between the trajectory data and the road network in the high-precision map, evaluate the rationality of the path and obtain a path rationality score;

[0025] Using the path features, behavior features and speed features, and the degree of compliance with traffic rules in the high-precision map, the behavior consistency is evaluated to obtain a behavior consistency score;

[0026] The evaluated location accuracy, path rationality and behavior consistency scores are taken as the evaluation results.

[0027] Optionally, a dynamic time warping algorithm is used to calculate the similarity between the trajectory data and the road network in the high-precision map, and the path rationality is evaluated to obtain a path rationality score including:

[0028] Using the DWT similarity calculation model, the similarity between the first point sequence formed by the trajectory data and the second point sequence corresponding to the road network in the high-precision map is calculated;

[0029] The first point trace sequence is sequentially subjected to geometric consistency evaluation, topological structure matching evaluation and lane-level precise matching evaluation to obtain an evaluation result, and the evaluation result is combined with the similarity to calculate a path rationality score.

[0030] Optionally, the step of sequentially performing geometric consistency evaluation, topological structure matching evaluation, and lane-level precise matching evaluation on the first point trace sequence to obtain an evaluation result, and calculating a path rationality score using the evaluation result in combination with the similarity includes:

[0031] Respectively calculating the lateral offset distance and the longitudinal path deviation between each track point in the first point track sequence and the center line of the road; calculating the lateral offset distance consistency index using the lateral offset distance and calculating the longitudinal path offset consistency index using the longitudinal path deviation; and using the lateral offset distance consistency index and the longitudinal path offset consistency index as the geometric consistency evaluation result;

[0032] Using the first point trace sequence, calculating a path continuity index, a turning angle deviation index, and a lane change rationality index, and using the three as topological structure matching evaluation results;

[0033] Calculating the lane matching degree using the first point trace sequence and using it as a lane-level accurate evaluation result;

[0034] The path rationality score is calculated using the similarity calculated between the first point trace sequence and the second point trace sequence, the geometric consistency evaluation results, the topological structure matching evaluation results and the lane-level precision evaluation results.

[0035] Optionally, the use of the path features, behavior features and speed features to determine the degree of compliance with traffic rules in the high-precision map to evaluate the behavior consistency and obtain a behavior consistency score includes:

[0036] Using the behavior characteristics, path characteristics, speed characteristics and the turning behavior, turning radius and speed in the traffic rules in the corresponding high-precision map, respectively perform turning behavior consistency evaluation, speed behavior consistency evaluation and speed behavior consistency evaluation results on the trajectory data to obtain a turning behavior consistency evaluation result, a turning radius consistency evaluation result and a speed behavior consistency evaluation result;

[0037] The behavior consistency score of the trajectory data is calculated using the turning behavior consistency evaluation result, the turning radius consistency evaluation result and the speed behavior consistency evaluation result.

[0038] Optionally, the use of the behavior characteristics, path characteristics, speed characteristics and the turning behavior, turning radius and speed in the traffic rules in the corresponding high-precision map to respectively perform a turning behavior consistency evaluation, a speed behavior consistency evaluation and a speed behavior consistency evaluation result on the trajectory data, and the obtained turning behavior consistency evaluation result, turning radius consistency evaluation result and speed behavior consistency evaluation result include:

[0039] Calculating a steering angle deviation using the behavior characteristics and the steering behavior in the traffic rules in the high-precision map, and calculating a steering consistency index using the steering angle deviation;

[0040] The turning radius consistency index is calculated by using the behavioral characteristics and the turning radius constraints in the traffic rules in the high-precision map;

[0041] Using the speed characteristics and the traffic rules acceleration and speed in the high-precision map, the acceleration consistency index, speed consistency index, lateral motion consistency index and lane change behavior consistency index are calculated;

[0042] A behavior consistency score is calculated by combining the steering consistency index, the turning radius consistency index, the acceleration consistency index, the speed consistency index, the lateral motion consistency index and the lane change behavior consistency index.

[0043] Optionally, the performing a comprehensive analysis on the evaluation results to obtain a quality analysis report, and feeding back the comprehensive analysis results to the roadside perception system includes:

[0044] Using a multi-layer attention deep neural network, comprehensively analyzing the scores in the evaluation results to obtain a comprehensive quality score;

[0045] Mark trajectory data with a comprehensive quality score lower than the preset threshold as abnormal data, and generate a quality analysis report containing the scores of various indicators and abnormal data;

[0046] The quality analysis report is fed back to the roadside perception system.

[0047] The innovative features of the embodiments of the present application include:

[0048] The embodiment of the present application discloses a real-time evaluation method for the quality of roadside perception data based on a high-precision map, by acquiring relevant data of a predetermined area; extracting features from the preprocessed data to obtain feature extraction results; using the feature extraction results to evaluate the quality of the trajectory data to obtain evaluation results; comprehensively analyzing the evaluation results to obtain a quality analysis report, and feeding back the comprehensive analysis results to the roadside perception system. The present application compares the perception data with the dynamic data in the high-precision map, not only focusing on the position accuracy perceived by the roadside perception system, but also comprehensively considering multi-dimensional indicators such as speed, path, and behavior, so as to conduct a comprehensive quality evaluation of the roadside perception system, thereby improving the accuracy and practicality of the evaluation results; and the present application establishes a real-time feedback mechanism to timely discover and correct problems in the trajectory data, and promote the continuous optimization of the roadside perception system. In addition, the present application adopts advanced algorithms and models to intelligently evaluate the quality of the trajectory data, thereby improving the evaluation efficiency and accuracy. Therefore, the present application can comprehensively and accurately evaluate the quality of the trajectory data in the roadside perception system, and promote the continuous optimization of the system through a real-time feedback mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art description are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can also be obtained based on these drawings without creative work.

[0050] Figure 1 A schematic diagram of a method for real-time evaluation of roadside perception data quality based on a high-precision map provided in an embodiment of the present application;

[0051] Figure 2 A principle block diagram of a method for real-time evaluation of roadside perception data quality based on a high-precision map provided in an embodiment of the present application. DETAILED DESCRIPTION

[0052] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0053] It should be noted that the terms "including" and "having" and any variations thereof in the embodiments of the present application and the accompanying drawings are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device comprising a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or devices.

[0054] The embodiment of the present application discloses a real-time evaluation method for the quality of roadside perception data based on a high-precision map. The method not only focuses on the position accuracy perceived by the roadside perception system, but also comprehensively considers multi-dimensional indicators such as speed, path, and behavior to conduct a comprehensive quality evaluation of the roadside perception system. The present application compares the perception data with the dynamic data in the high-precision map, which improves the accuracy and practicality of the evaluation results; and the present application establishes a real-time feedback mechanism to promptly discover and correct problems in the trajectory data, and promote the continuous optimization of the roadside perception system. In addition, the present application uses advanced algorithms and models to conduct a comprehensive and intelligent evaluation of the quality of trajectory data, thereby improving the efficiency and accuracy of the evaluation. Therefore, the present application can comprehensively and accurately evaluate the quality of trajectory data in the roadside perception system, and promote the continuous optimization of the system through a real-time feedback mechanism.

[0055] The embodiments of the present application are described in detail below.

[0056] Combination Figure 1 and Figure 2 The present application provides a method for real-time evaluation of roadside perception data quality based on a high-precision map, including:

[0057] S1, obtaining relevant data of a predetermined area, wherein the relevant data includes dynamic data of a high-precision map and trajectory data of traffic participants in the predetermined area obtained from a roadside perception system;

[0058] The relevant data collection of this application is divided into two aspects. The first aspect is to obtain the trajectory data of traffic participants from the roadside perception system, including location, speed, acceleration, heading and other information. The second aspect is to obtain static and dynamic information such as road geometry, lane information, traffic signs, and signal lights in high-precision maps.

[0059] S2, performing data preprocessing on the relevant data to obtain preprocessed data;

[0060] S3, performing feature extraction on the preprocessed data to obtain a feature extraction result;

[0061] refer to Figure 2 , the feature extraction results include: position features, speed features, path features and behavior features; the position features represent the position coordinates of traffic participants at different time points, the speed features represent the instantaneous speed and average speed of traffic participants, the path features represent the driving paths of traffic participants, and the behavior features represent the driving behaviors of traffic participants. The speed features include the instantaneous speed and average speed of traffic participants, the path features include the driving paths of traffic participants, including turning, going straight and changing lanes, etc.; the behavior features include the driving behaviors of traffic participants, such as acceleration, deceleration and parking, etc.

[0062] S4, using the feature extraction result to perform quality assessment on the trajectory data to obtain an assessment result;

[0063] In this step, the feature extraction results can be used to respectively calculate the position accuracy score, the path rationality score and the behavior consistency score, and the three are used as the evaluation results of the trajectory data.

[0064] S5, comprehensively analyzing the evaluation results to obtain comprehensive analysis results, and feeding back the comprehensive analysis results to the roadside perception system.

[0065] In a specific embodiment of the present application, reference Figure 2 , the preprocessing of the relevant data to obtain preprocessed data includes:

[0066] a1, performing data cleaning on the relevant data to remove missing, abnormal or erroneous data points to obtain cleaned data;

[0067] This step eliminates missing, abnormal or erroneous data points to ensure data quality and provide accurate data basis for subsequent quality evaluation.

[0068] a2, synchronizing the timestamps of the cleaned data to obtain synchronized data;

[0069] This step synchronizes the timestamps of the relevant data perceived by the roadside perception system with the timestamps of the relevant data obtained from the high-precision map, ensuring that the data are compared on the same time scale.

[0070] a3, converting the coordinate system of the synchronized data to obtain data in a unified coordinate system, and using the data as preprocessed data.

[0071] This step performs time-space alignment, converting the coordinate system of the relevant data perceived by the roadside perception system and the data obtained by the high-precision map to ensure consistency in spatial position.

[0072] In a specific implementation of the present application, the use of the feature extraction results to respectively calculate the position accuracy score, the path rationality score and the behavior consistency score, and using the three as the evaluation results of the trajectory data includes:

[0073] b1, evaluating the position accuracy by using the Euclidean distance between the first position coordinate represented by the position feature and the corresponding second position coordinate in the high-precision map;

[0074] The Euclidean distance is expressed as:

[0075]

[0076] The position accuracy is expressed as:

[0077] Accuracy_Pos=1-(D / Dmax)

[0078] Where Dmax is the preset maximum allowable deviation distance. When D is less than Dmax, Accuracy_Pos is close to 1, indicating high position accuracy; otherwise, it is close to 0.

[0079] b2, using the dynamic time warping algorithm, calculate the similarity between the trajectory data and the road network in the high-precision map, evaluate the rationality of the path and obtain a path rationality score;

[0080] The path rationality assessment aims to systematically verify whether the trajectory data conforms to the road network logic in the high-precision map, and ensure the rationality of the driving path of traffic participants in terms of space and topology. This application uses the Dynamic Time Warping (DTW) algorithm as the core matching technology, combined with multi-dimensional path feature analysis, to comprehensively evaluate the rationality of the path.

[0081] b3, using the path characteristics, behavior characteristics and speed characteristics, and the degree of compliance with the traffic rules in the high-precision map, to evaluate the behavior consistency and obtain a behavior consistency score;

[0082] Behavior consistency assessment aims to systematically verify whether the kinematic behavior of traffic participants in trajectory data conforms to the traffic rules and road characteristics in high-precision maps, ensuring that the perception data reflects real and reasonable traffic behavior.

[0083] b4, the evaluated location accuracy, path rationality and behavior consistency scores are used as evaluation results.

[0084] In a specific implementation of the present application, a dynamic time warping algorithm is used to calculate the similarity between the trajectory data and the road network in the high-precision map, and the rationality of the path is evaluated to obtain a reasonable path score, including:

[0085] b21, using the DWT similarity calculation model, calculate the similarity between the first point sequence formed by the trajectory data and the second point sequence corresponding to the road network in the high-precision map; the DWT similarity calculation model is expressed as:

[0086] DTW(P,Q)=min{∑d(pi,qi)|pi∈P,qi∈Q}

[0087] Where P is the first point sequence formed by the trajectory data, Q is the corresponding second point sequence in the road network, and d(pi,qi) represents the distance measure between the two points;

[0088] b22, performing geometric consistency evaluation, topological structure matching evaluation and lane-level precise matching evaluation on the first point trace sequence in sequence to obtain an evaluation result, and using the evaluation result in combination with the similarity to calculate a path rationality score.

[0089] In a specific implementation of the present application, the step of sequentially performing geometric consistency evaluation, topological structure matching evaluation, and lane-level precise matching evaluation on the first point trace sequence to obtain an evaluation result, and using the evaluation result in combination with the similarity to calculate the path rationality score includes:

[0090] b221, respectively calculating the lateral offset distance and the longitudinal path deviation between each track point in the first point track sequence and the center line of the road; calculating the lateral offset distance consistency index using the lateral offset distance and calculating the longitudinal path offset consistency index using the longitudinal path deviation; and using the lateral offset distance consistency index and the longitudinal path offset consistency index as the geometric consistency evaluation result;

[0091] The lateral offset distance is expressed as: D_lateral(pi) = |d(pi,L)|, and the lateral offset distance consistency index is expressed as:

[0092] Consistency_Lateral=1-(∑D_lateral(pi) / n) / D_max_lateral

[0093] Where n is the number of trajectory points, D_max_lateral is the preset maximum allowable lateral offset distance, pi is the trajectory point, L is the road centerline, and d(pi,L) represents the vertical distance from the trajectory point to the road centerline;

[0094] The longitudinal path deviation is expressed as: D_longitudinal = |L_actual-L_expected|, and the longitudinal path deviation consistency index is expressed as:

[0095] Consistency_Longitudinal=1-(D_longitudinal / L_expected)

[0096] In the formula, L_actual is the actual driving path length, and L_expected is the optimal path length in the high-precision map;

[0097] b222, using the first point trace sequence, calculating a path continuity index, a turning angle deviation index, and a lane change rationality index, and using the three as topology structure matching evaluation results;

[0098] Wherein, the path continuity index is expressed as:

[0099] S_topology = 1-(number of breakpoints / total number of path points)

[0100] The turning angle deviation index is expressed as:

[0101] Δθ=|θactual-θexpected|

[0102] In the formula, θactual represents the actual steering angle, and θexpected represents the expected steering angle predicted in the high-precision map;

[0103] The lane change rationality index is expressed as:

[0104] S_lane_change = 1-(number of lane changes / path length)

[0105] b223, using the first point trace sequence to calculate the lane matching degree, and use it as the lane level accurate evaluation result; the lane matching degree is expressed as:

[0106] S_lane = 1-(number of lane switches / total path length)

[0107] b224, using the similarity calculated between the first point trace sequence and the second point trace sequence, the geometric consistency evaluation result, the topological structure matching evaluation result and the lane-level precision evaluation result, calculate the path rationality score. The path rationality score is expressed as:

[0108] S_path=w11*Consistency_Lateral

[0109] +w21*Consistency_Longitudinal

[0110] +w31*S_topology

[0111] +w41*S_lane_change

[0112] +w51*S_lane

[0113] In the formula, w11, w21, w31, w41, and w51 are weight coefficients, and their sum is equal to 1.

[0114] In a specific implementation of the present application, the use of the path features, behavior features and speed features to determine the degree of compliance with traffic rules in the high-precision map to evaluate the behavior consistency and obtain a behavior consistency score includes:

[0115] b31, using the behavior characteristics, path characteristics, speed characteristics and the turning behavior, turning radius and speed in the traffic rules in the corresponding high-precision map, respectively perform turning behavior consistency evaluation, speed behavior consistency evaluation and speed behavior consistency evaluation results on the trajectory data to obtain a turning behavior consistency evaluation result, a turning radius consistency evaluation result and a speed behavior consistency evaluation result;

[0116] b32, using the steering behavior consistency evaluation result, the turning radius consistency evaluation result and the speed behavior consistency evaluation result, calculate the behavior consistency score of the trajectory data.

[0117] In a specific implementation of the present application, the behavior characteristics, path characteristics, speed characteristics and the turning behavior, turning radius and speed in the traffic rules in the corresponding high-precision map are used to perform turning behavior consistency evaluation, speed behavior consistency evaluation and speed behavior consistency evaluation results on the trajectory data, respectively, and the turning behavior consistency evaluation results, turning radius consistency evaluation results and speed behavior consistency evaluation results are obtained, including:

[0118] b311, calculating a steering angle deviation using the behavior characteristics and the steering behavior in the traffic rules in the high-precision map, and calculating a steering consistency index using the steering angle deviation;

[0119] Among them, the steering angle deviation is expressed as:

[0120] Δθ=|θactual-θexpected|

[0121] Where θactual is the actual steering angle, and θexpected is the expected steering angle predicted by the high-precision map;

[0122] The steering consistency index is expressed as:

[0123] S_turn = 1 - (Δθ / θmax)

[0124] Where θmax is the preset maximum allowable angle deviation;

[0125] b312, using the behavior characteristics and the turning radius constraints in the traffic rules in the high-precision map, calculates the turning radius consistency index;

[0126] The turning radius is expressed as:

[0127] R_consistency=|Ractual-Rexpected| / Rexpected

[0128] The turning radius consistency index is expressed as:

[0129] S_radius = 1-R_consistency

[0130] In the formula, Ractual represents the actual turning radius, and Rexpected represents the expected turning radius predicted by the high-precision map;

[0131] b313, using the speed features and the acceleration and speed of the traffic rules in the high-precision map, calculate the acceleration consistency index, speed consistency index, lateral motion consistency index and lane change behavior consistency index;

[0132] The acceleration in the velocity characteristic is expressed as: Jerk = |d 2 a / dt 2 |; The acceleration consistency index is expressed as:

[0133] S_jerk=1-(Jerk / Jerk_max)

[0134] Wherein, Jerk_max represents the preset maximum acceleration threshold;

[0135] The speed consistency index is expressed as: S_speed = 1-|Vactual-Vlimit| / Vlimit

[0136] In the formula, Vactual represents the actual speed, and Vlimit represents the road speed limit;

[0137] The lateral acceleration in the velocity characteristic is expressed as: a_lateral = v 2 / R, v represents the lateral velocity, R represents the lateral radius;

[0138] The lateral motion consistency index is expressed as:

[0139]

[0140] The lane change angle of the lane change behavior in the behavior feature is expressed as: Δλ = |λactual-λexpected|, where λactual is the actual lane change angle and λexpected is the expected lane change angle predicted by the high-precision map; the lane change behavior consistency index is expressed as:

[0141] S_lane_change=1-(Δλ / λmax);

[0142] Where λmax is the maximum lane change angle threshold;

[0143] b314, calculating a behavior consistency score by combining the steering consistency index, the steering radius consistency index, the acceleration consistency index, the speed consistency index, the lateral motion consistency index and the lane change behavior consistency index. The behavior consistency score is expressed as:

[0144] S_behavior=w12*S_turn+w22*S_radius+w32*S_jerk+w42*S_speed+w52*S_lateral+w62*S_lane_change

[0145] In the formula, w12, w22, w32, w42, w52, and w62 are weight coefficients.

[0146] In a specific implementation of the present application, the comprehensive analysis of the evaluation results to obtain a quality analysis report, and feeding back the comprehensive analysis results to the roadside perception system includes:

[0147] c1, using a multi-layer attention deep neural network to comprehensively analyze the scores in the evaluation results to obtain a comprehensive quality score;

[0148] This application uses a multi-level attention deep neural network (Multi-Level Attention Deep Neural Network, MLA-DNN) to perform a comprehensive quality score on the trajectory data. The network structure of the multi-level attention deep neural network includes: an input layer for inputting data, which is the evaluated position accuracy, path rationality and behavior consistency score; a multi-head attention feature extraction layer for multi-dimensional feature vector extraction; a multi-task learning layer for learning multi-dimensional features; a comprehensive score output layer for outputting a comprehensive quality score. The multi-head attention feature extraction layer of this application adopts a multi-head attention mechanism, which can adaptively extract key features; and in the process of extracting key features, attention weights are assigned to corresponding features through attention weight calculation; the attention weight calculation formula is expressed as:

[0149] Attention(Q,K,V)=softmax(QK T / d k )V

[0150] Where Q represents the query matrix, K represents the key matrix, V represents the value matrix, and d k represents the dimension of the key vector, and the superscript T represents the transposition of the matrix;

[0151] The multi-layer attention deep neural network is trained in advance, and the multi-task learning loss function is used to optimize the training process during the training process. The multi-task learning loss function includes position accuracy loss, position accuracy loss L_positio, n-path rationality loss L_path, behavior consistency loss L_behavior and regularization term L_regularization; the multi-task learning loss function is expressed as:

[0152] L_total=L_position+L_path+L_behavior+λ*L_regularization, where λ is the regularization coefficient;

[0153] The output of the multi-layer attention deep neural network is a comprehensive quality score in the interval [0, 1]. Since the processing process of the multi-layer attention deep neural network is relatively mature, this application will not be described in detail here.

[0154] c2. Mark the trajectory data with a comprehensive quality score lower than the preset threshold as abnormal data, and generate a quality analysis report including the scores of various indicators and the abnormal data;

[0155] c3. Feed back the quality analysis report to the roadside perception system.

[0156] The present application takes the following feedback measures based on the comprehensive quality score:

[0157] Set the scoring threshold: According to the system requirements, set the scoring threshold y_thresh. Based on the comprehensive quality score y(t) being lower than this threshold, corresponding measures are executed. If it is lower than this threshold, the feedback strategy is executed. If it is not lower than this threshold, it operates normally and continuously monitors the trajectory data.

[0158] (1) When the comprehensive quality score y(t) is lower than this threshold, it indicates that the quality of the perception data is poor, that is, y(t) < y_thresh, then the following feedback strategy is executed:

[0159] Alarm prompt: Send an alarm to the driver or the monitoring center to prompt that there are quality problems with the trajectory data.

[0160] System adjustment: Adjust the parameters or strategies of the roadside perception system, such as increasing the sensor weight, adjusting the algorithm parameters, etc., to improve the data quality.

[0161] (2) If the quality of the trajectory data is good, y(t) > y_thresh, then the following strategy is executed:

[0162] Normal operation: Continue to use the current perception data for decision-making and control.

[0163] Performance monitoring: Continuously monitor the quality of the perception data to ensure the stable operation of the system.

[0164] The embodiment of the present application discloses a real-time evaluation method for the quality of roadside perception data based on a high-precision map. This method not only focuses on the position accuracy perceived by the roadside perception system, but also comprehensively considers multi-dimensional indicators such as speed, path, and behavior to conduct a comprehensive quality assessment of the roadside perception system. The present application compares the perception data with the dynamic data in the high-precision map, improving the accuracy and practicality of the evaluation results; and the present application establishes a real-time feedback mechanism to timely discover and correct problems in the trajectory data, promoting the continuous optimization of the roadside perception system. In addition, the present application adopts advanced algorithms and models to comprehensively and intelligently evaluate the quality of the trajectory data, improving the evaluation efficiency and accuracy. Therefore, the present application can comprehensively and accurately evaluate the quality of the trajectory data in the roadside perception system and promote the continuous optimization of the system through the real-time feedback mechanism.

[0165] Those skilled in the art can understand that the accompanying drawings are only schematic diagrams of one embodiment, and the modules or processes in the accompanying drawings are not necessarily required to implement the present application.

[0166] Those skilled in the art can understand that the modules in the device in the embodiment can be distributed in the device in the embodiment according to the description of the embodiment, or can be changed accordingly and located in one or more devices different from the embodiment. The modules in the above embodiment can be combined into one module, or can be further divided into multiple sub-modules.

[0167] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A real-time evaluation method for roadside perception data quality based on high-precision maps, characterized in that: include: Acquire relevant data of a predetermined area, wherein the relevant data includes dynamic data of a high-precision map and trajectory data of traffic participants in the predetermined area acquired from a roadside perception system; Performing data preprocessing on the relevant data to obtain preprocessed data; Performing feature extraction on the preprocessed data to obtain a feature extraction result; Using the feature extraction result to perform quality assessment on the trajectory data to obtain an assessment result; The evaluation results are comprehensively analyzed to obtain a comprehensive analysis result, and the comprehensive analysis result is fed back to the roadside perception system.

2. The method for real-time evaluation of roadside perception data quality based on high-precision maps according to claim 1 is characterized in that: The performing data preprocessing on the relevant data to obtain preprocessed data comprises: Performing data cleaning on the relevant data to remove missing, abnormal or erroneous data points to obtain cleaned data; Synchronizing the timestamps of the cleaned data to obtain synchronized data; The coordinate system of the synchronized data is converted to obtain data in a unified coordinate system, and the data is used as preprocessed data.

3. The method for real-time evaluation of roadside perception data quality based on high-precision maps according to claim 1 is characterized in that: The feature extraction results include: position features, speed features, path features and behavior features; the position features represent the position coordinates of traffic participants at different time points, the speed features represent the instantaneous speed and average speed of traffic participants, the path features represent the driving paths of traffic participants, and the behavior features represent the driving behaviors of traffic participants.

4. The method for real-time evaluation of roadside perception data quality based on high-precision maps according to claim 3 is characterized in that: The using the feature extraction result to perform quality assessment on the trajectory data to obtain an assessment result comprises: The feature extraction results are used to calculate the position accuracy score, path rationality score and behavior consistency score respectively, and the three are used as the evaluation results of the trajectory data.

5. The method for real-time evaluation of roadside perception data quality based on high-precision maps according to claim 4 is characterized in that: The feature extraction results are used to respectively calculate the position accuracy score, the path rationality score and the behavior consistency score, and the three are used as the evaluation results of the trajectory data, including: Evaluating the position accuracy by using the Euclidean distance between the first position coordinate represented by the position feature and the corresponding second position coordinate in the high-precision map; Use the dynamic time warping algorithm to calculate the similarity between the trajectory data and the road network in the high-precision map, evaluate the rationality of the path and obtain a path rationality score; Using the path features, behavior features and speed features, and the degree of compliance with traffic rules in the high-precision map, the behavior consistency is evaluated to obtain a behavior consistency score; The evaluated location accuracy, path rationality and behavior consistency scores are taken as the evaluation results.

6. The method for real-time evaluation of roadside perception data quality based on high-precision maps according to claim 5 is characterized in that: The dynamic time warping algorithm is used to calculate the similarity between the trajectory data and the road network in the high-precision map, and the rationality of the path is evaluated to obtain the path rationality score including: Using the DWT similarity calculation model, the similarity between the first point sequence formed by the trajectory data and the second point sequence corresponding to the road network in the high-precision map is calculated; The first point trace sequence is sequentially subjected to geometric consistency evaluation, topological structure matching evaluation and lane-level precise matching evaluation to obtain an evaluation result, and the evaluation result is combined with the similarity to calculate a path rationality score.

7. The method for real-time evaluation of roadside perception data quality based on high-precision maps according to claim 6 is characterized in that: The step of sequentially performing geometric consistency evaluation, topological structure matching evaluation, and lane-level precise matching evaluation on the first point trace sequence to obtain an evaluation result, and using the evaluation result in combination with the similarity to calculate a path rationality score includes: Respectively calculating the lateral offset distance and the longitudinal path deviation between each track point in the first point track sequence and the center line of the road; calculating the lateral offset distance consistency index using the lateral offset distance and calculating the longitudinal path offset consistency index using the longitudinal path deviation; and using the lateral offset distance consistency index and the longitudinal path offset consistency index as the geometric consistency evaluation result; Using the first point trace sequence, calculating a path continuity index, a turning angle deviation index, and a lane change rationality index, and using the three as topological structure matching evaluation results; Calculating the lane matching degree using the first point trace sequence and using it as a lane-level accurate evaluation result; The path rationality score is calculated using the similarity calculated between the first point trace sequence and the second point trace sequence, the geometric consistency evaluation results, the topological structure matching evaluation results and the lane-level precision evaluation results.

8. The method for real-time evaluation of roadside perception data quality based on high-precision maps according to claim 5 is characterized in that: The use of the path features, behavior features and speed features to determine the degree of compliance with traffic rules in the high-precision map to evaluate the behavior consistency and obtain a behavior consistency score includes: Using the behavior characteristics, path characteristics, speed characteristics and the turning behavior, turning radius and speed in the traffic rules in the corresponding high-precision map, respectively perform turning behavior consistency evaluation, speed behavior consistency evaluation and speed behavior consistency evaluation results on the trajectory data to obtain a turning behavior consistency evaluation result, a turning radius consistency evaluation result and a speed behavior consistency evaluation result; The behavior consistency score of the trajectory data is calculated using the turning behavior consistency evaluation result, the turning radius consistency evaluation result and the speed behavior consistency evaluation result.

9. The method for real-time evaluation of roadside perception data quality based on high-precision maps according to claim 8 is characterized in that: The turning behavior consistency evaluation, speed behavior consistency evaluation and speed behavior consistency evaluation result are respectively performed on the trajectory data by using the behavior characteristics, path characteristics, speed characteristics and the turning behavior, turning radius and speed in the traffic rules in the corresponding high-precision map, and the turning behavior consistency evaluation result, turning radius consistency evaluation result and speed behavior consistency evaluation result are obtained, including: Calculating a steering angle deviation using the behavior characteristics and the steering behavior in the traffic rules in the high-precision map, and calculating a steering consistency index using the steering angle deviation; The turning radius consistency index is calculated by using the behavioral characteristics and the turning radius constraints in the traffic rules in the high-precision map; Using the speed characteristics and the traffic rules acceleration and speed in the high-precision map, the acceleration consistency index, speed consistency index, lateral motion consistency index and lane change behavior consistency index are calculated; A behavior consistency score is calculated by combining the steering consistency index, the turning radius consistency index, the acceleration consistency index, the speed consistency index, the lateral motion consistency index and the lane change behavior consistency index.

10. The method for real-time evaluation of roadside perception data quality based on high-precision maps according to claim 1, characterized in that: The performing comprehensive analysis on the evaluation results to obtain a quality analysis report, and feeding back the comprehensive analysis results to the roadside perception system comprises: Using a multi-layer attention deep neural network, comprehensively analyzing the scores in the evaluation results to obtain a comprehensive quality score; Mark trajectory data with a comprehensive quality score lower than the preset threshold as abnormal data, and generate a quality analysis report containing the scores of various indicators and abnormal data; The quality analysis report is fed back to the roadside perception system.

Images