Low-speed ungrouped vehicle identification system

Abstract

The invention provides a low-speed ungrouped vehicle identification system. According to the system, a heavy vehicle basic information statistics module obtains basic information and driving data of a vehicle and front and rear vehicles; the characteristic index module calculates a characteristic index for judgment; the low-speed ungrouped vehicle judgment module constructs a fuzzy comprehensive evaluation model according to the characteristic indexes, judges whether the vehicles belong to the low-speed ungrouped vehicles or not, and constructs labels for the vehicles; the recognition model training module takes a vehicle as a unit, and constructs a vehicle information data set based on the driving behavior basic information and labels of the vehicle; based on the vehicle information data set, the low-speed ungroup identification model is trained, so that the model can judge whether the target vehicle belongs to the low-speed ungroup vehicle according to the basic information and driving data of the target vehicle and the front and rear vehicles, equipment safety and a large amount of cost investment are avoided, and misjudgment or missed judgment is prevented.

Description

Technical Field

[0001] The present invention relates to the field of intelligent traffic monitoring, and particularly to a system for identifying low-speed unsociable vehicles. Background Art

[0002] Low-speed unsociable vehicles will block the vehicles behind. The blocked vehicles will increase the speed discreteness during acceleration, deceleration and lane-changing processes, which will affect the traffic safety of the road section. This kind of influence continues to spread upstream and is also prone to form a moving bottleneck, resulting in traffic congestion. Therefore, it becomes very necessary to effectively manage and control low-speed unsociable vehicles.

[0003] Currently, the main low-speed detection method in China is to compare the vehicle speed with the section speed limit and the average speed to determine whether the vehicle is a low-speed vehicle. Common detection devices include radars, inductive loops, etc. Radars have high costs and limited accuracy for high-density traffic flows. Inductive loops require equipment to be buried and have high maintenance costs.

[0004] Both the above methods and devices have certain problems. On the one hand, relying solely on speed as a judgment index is too single and prone to misjudgment or missed judgment. On the other hand, the costs of the devices are relatively high. Summary of the Invention

[0005] The purpose of this application is to provide a system for identifying low-speed unsociable vehicles, a method, a method for training a low-speed unsociable vehicle identification model, and an electronic device, which are used to solve or overcome the above problems existing in the prior art.

[0006] According to the first aspect of this application, a system for identifying low-speed unsociable vehicles is provided, which includes: a vehicle basic information statistics module, a characteristic index module, a low-speed unsociable determination module, and an identification model training module;

[0007] The vehicle basic information statistics module is used to determine the basic information and driving data of the vehicle and the vehicle in front and behind based on the following processing process: analyze the vehicles within the field of view, obtain the driving data of the vehicle and the vehicles in front and behind, and use it as the basic information of the driving behavior of the vehicle at the initial stage.

[0008] The characteristic index module is used to obtain the basic information of the vehicle driving behavior based on the following processing: calculate the information obtained in the vehicle basic information statistics module, and obtain the characteristic index data corresponding to each vehicle from the speed, acceleration, speed difference from the vehicle in front, acceleration difference, distance difference, speed difference from the vehicle behind, acceleration difference, and distance difference.

[0009] The low-speed unsociable determination module is used to determine whether the vehicle belongs to a low-speed unsociable vehicle based on the following processing:

[0010] Determine its threshold and weight according to the characteristic index data;

[0011] Determine the membership function according to the threshold and weight, construct a fuzzy comprehensive evaluation model, judge whether each vehicle belongs to a low-speed unsociable vehicle, and construct labels.

[0012] The recognition model training module is used to train the low-speed unsociable vehicle recognition model based on the following processing:

[0013] Taking the vehicle as a unit, construct a vehicle information data set based on the basic driving behavior information and labels of the corresponding vehicle;

[0014] Based on the vehicle information data set, train the low-speed unsociable vehicle recognition model so that the model can judge whether the target vehicle belongs to a low-speed unsociable vehicle according to the basic driving behavior information of the target vehicle.

[0015] A method for identifying low-speed unsociable vehicles, which includes:

[0016] Obtain the driving behavior information of the target vehicle and the vehicles in front and behind, and summarize it as the basic driving behavior information of the target vehicle;

[0017] Based on the basic driving behavior information of the target vehicle, calculate the speed, acceleration, and position of the vehicle relative to the vehicles in front and behind, and summarize them with the vehicle's own driving behavior data as the characteristic indicators of the target vehicle;

[0018] Construct a membership function based on the characteristic indicators of the target vehicle, and determine whether the target vehicle belongs to a low-speed unsociable vehicle according to the total membership result;

[0019] Input the basic driving behavior information obtained by monitoring and the determination label obtained by calculating the characteristic indicators of the basic driving behavior information into the recognition model to train the model.

[0020] A method for training a low-speed unsociable vehicle recognition model, which includes:

[0021] Correspond the obtained basic driving behavior information of the vehicle and the label of the vehicle;

[0022] Construct a vehicle information data set according to the basic driving behavior information of the vehicle and the vehicle label;

[0023] Based on the vehicle information data set, train the low-speed unsociable vehicle recognition model so that the model can judge whether the target vehicle belongs to a low-speed unsociable vehicle according to the basic driving information of the target vehicle.

[0024] An electronic device, characterized in that it includes a memory and a processor to implement:

[0025] Based on the vehicle trajectory video obtained by monitoring, through vehicle detection and tracking, convert pixel coordinates into actual coordinates, and finally export the data of each vehicle;

[0026] Calculate the characteristic values according to the exported vehicle data, construct the membership function, and determine the vehicle label;

[0027] Taking the vehicle as a unit, construct a vehicle information data set according to the corresponding basic driving information and vehicle label of the vehicle;

[0028] Based on the vehicle information data set, train the low-speed unsociable vehicle recognition model, so that the model can judge whether the target vehicle belongs to the low-speed unsociable vehicle according to the basic driving information of the target vehicle. Description of the Drawings

[0029] Figure 1 It is a schematic structural diagram of a low-speed unsociable vehicle recognition system according to an embodiment of the present application.

[0030] Figure 2 It is the real-time export of vehicle data from the monitoring perspective.

[0031] Figure 3 It is the membership function and image. Detailed Embodiment

[0032] In order to enable those skilled in the art to better understand the technical solutions in the embodiments of the present application, the following will clearly and completely describe the technical solutions in the embodiments of the present application in conjunction with the accompanying drawings in the embodiments of the present application.

[0033] Figure 1 It is a schematic structural diagram of a low-speed unsociable vehicle recognition system according to an embodiment of the present application, which includes a vehicle basic information statistics module, a characteristic index module, a low-speed unsociable determination module, and an identification model training module;

[0034] The vehicle basic information statistics module is used to determine the basic information and driving data of the vehicle and the vehicle in front and behind based on the following processing process: analyze the vehicles within the field of view, obtain the driving data of the vehicle and the vehicle in front and behind, and use it as the basic information of the driving behavior of the vehicle in the initial stage;

[0035] The characteristic index module is used to obtain the basic information of the vehicle driving behavior based on the following processing: calculate the information obtained in the vehicle basic information statistics module, and obtain the corresponding characteristic index data of each vehicle from the speed, acceleration, speed difference from the vehicle in front, acceleration difference, distance difference, speed difference from the vehicle behind, acceleration difference, and distance difference;

[0036] The low-speed unsociable determination module is used to determine whether the vehicle belongs to the low-speed unsociable vehicle based on the following processing:

[0037] Determine its threshold and weight according to the characteristic index data;

[0038] Determine the membership function according to the threshold and weight, construct a fuzzy comprehensive evaluation model, judge whether each vehicle belongs to a low-speed unsociable vehicle, and construct labels;

[0039] The recognition model training module is used to train the low-speed unsociable vehicle recognition model based on the following processing:

[0040] Taking the vehicle as a unit, construct a vehicle information data set based on the basic driving behavior information and labels of the corresponding vehicle;

[0041] Based on the vehicle basic information data set, train the low-speed unsociable vehicle recognition model, so that the model can judge whether the target vehicle belongs to a low-speed unsociable vehicle according to the basic driving behavior information of the target vehicle. The basic driving behavior information of the vehicle is input into the model in the form of a time series, and different time lengths are divided through a time window to predict whether the vehicle belongs to a low-speed unsociable vehicle. For the input data, the model iterates continuously to find the parameter values that minimize the objective function, obtain the optimal solution, and thus realize that for new vehicle samples, the model can identify and predict whether the vehicle belongs to a low-speed unsociable vehicle according to the input data.

[0042] The above recognition model can be a machine learning model, such as Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), Random Forest (BF). Among them, Extreme Gradient Boosting (XGBoost) uses the second-order Taylor expansion of the loss function as a surrogate function, and determines the best splitting point and leaf node output value of the regression tree by solving the derivative equal to 0 to identify and predict whether the vehicle belongs to a low-speed unsociable vehicle; Support Vector Machine (SVM) finds an optimal hyperplane that meets the classification requirements while ensuring the classification accuracy, makes the distance between the hyperplane and the points of the two types of samples the largest, and introduces a kernel function k(x i ,x j ), map the vehicle sample data {(x i ,y i )} to a high-dimensional space, transform it into a linearly separable problem in the high-dimensional space, and finally construct the optimal hyperplane in the high-dimensional space to identify and predict whether the vehicle belongs to a low-speed unsociable vehicle; Random Forest (BF) constructs multiple decision trees through Bagging and random feature selection, and obtains the final predicted category of the vehicle through voting.

[0043] Figure 2For the real-time export of vehicle data from the monitoring perspective. For the data in the vehicle basic information statistics module, target vehicle detection can be performed using YOLOv8 based on the monitoring video, combined with a vehicle tracking algorithm (such as DeepSORT), assign a unique ID to each vehicle, continuously track it in the video, then convert the pixel coordinates to actual coordinates through perspective transformation, calculate the vehicle speed, acceleration, and position, and finally export the trajectory data of all vehicles and surrounding vehicles to form the basic driving information of the vehicle.

[0044] Taking the vehicle as the unit, preprocess based on the corresponding basic driving information of the vehicle. Optionally, the preprocessing includes but is not limited to filling missing values, removing duplicate values, outliers, etc., to ensure the continuity and consistency of the data.

[0045] Based on the preprocessed basic driving information of the vehicle, calculate the driving feature indicators, specifically including: speed, acceleration, speed difference from the vehicle in front, acceleration difference from the vehicle in front, distance difference from the vehicle in front, speed difference from the vehicle behind, acceleration difference from the vehicle behind, distance difference from the vehicle behind.

[0046] Based on the vehicle driving feature indicators, use the interquartile range method to determine the threshold of each feature indicator. The formula is as follows:

[0047] M i =Q i ±1.5 I i

[0048] In the formula: M i is the threshold of the i-th low-speed outlier driving behavior feature indicator; Q i When it is the upper quartile value, the sign takes "+"; Q i When it is the lower quartile value, the sign takes "-"; I i is the difference between the upper and lower quartile values.

[0049] Use CRITIC to determine the weight of each feature indicator. The formula is as follows:

[0050]

[0051] In the formula: w i is the weight of the i-th feature indicator; c i is the information amount of the i-th feature indicator; δ i is the standard deviation of the i-th feature indicator, that is, the comparison intensity of the feature indicator; r ij is the correlation coefficient between feature indicator i and feature indicator j.

[0052] Adopt the fuzzy comprehensive evaluation method to obtain the vehicle label. The specific steps are as follows:

[0053] Determine the evaluation factor set. The factor set is a set composed of various factors that affect the evaluation object, U = {u1, u2, u3, …, u i , … u n}, where the evaluation factors are the above 8 indicators;

[0054] Determine the evaluation set. The evaluation set is a set composed of various possible results that the evaluation object belongs to, V = {v1, v2, v3, …, v j , … v m}, where the evaluation set is divided into low-speed unsociable type, normal type, and overspeed type;

[0055] Determine the fuzzy comprehensive evaluation matrix. The fuzzy comprehensive judgment matrix is as follows:

[0056]

[0057] In the formula, R ij is the membership degree of the i-th vehicle with respect to the j-th evaluation set.

[0058] Judgment. According to the magnitudes of the membership degrees of each evaluation set, determine which evaluation set the sample belongs to. Here, it is to determine whether the vehicle belongs to low-speed unsociable, normal, or overspeed.

[0059] The membership function satisfies 0 <= u(x) <= 1, that is, for each element x, there is a corresponding value. In practical applications, trapezoidal membership functions and triangular membership functions are often used. Combining with the analysis of the actual application scenario, determine which type of the function belongs to, i.e., the partial large type, partial small type, and intermediate type, and determine the values of each segmentation point. Optionally, choose to construct a trapezoidal membership function, and the function and graph are as Figure 3 shown.

[0060] The partial small type means that for the x to be evaluated, the smaller its value, the higher the corresponding membership degree, and the greater the possibility of belonging to this type; the intermediate type means that for the x to be evaluated, when its value belongs to the intermediate range, the higher the corresponding membership degree, and the greater the possibility of belonging to this type; the partial large type means that for the x to be evaluated, the larger its value, the higher the corresponding membership degree, and the greater the possibility of belonging to this type. Combining with the actual situation of the indicators in each evaluation set, judge the type of its membership function, and the values of each segmentation point of the function are as follows in the table:

[0061]

[0062] Combined with the weights, calculate the total membership degrees of the 8 characteristic indicators in each evaluation set, and compare them. The evaluation set with the largest total membership degree is the type to which the vehicle belongs.

[0063] Construct a vehicle information data set based on the basic driving data and labels of the vehicle, and use the SMOTE method, the ADASYN method, and the method of adjusting the scale pos weight parameter to perform sample balancing processing on the vehicle information data set.

[0064] Based on the vehicle information data set, the recognition model training module trains the recognition model and performs the following steps based on machine learning models such as Extreme Gradient Boosting Tree (XGBoost), Support Vector Machine (SVM), and Random Forest (BF):

[0065] Divide the samples into a training set and a test set according to a ratio of 4:1;

[0066] Data standardization;

[0067] Parameter optimization and evaluation of model performance.

[0068] When evaluating the performance of the model, the model training module performs the following steps:

[0069] Count the proportion of the number of correct samples in all samples to the total number of samples to obtain the accuracy rate, which is used to represent the overall recognition effect of the model;

[0070] Count the proportion of the data of the same type in all samples whose true results also belong to this type to obtain the precision rate;

[0071] Count the proportion of the data of the same class in all samples that are recognized and classified as this class by the recognition model to obtain the recall rate;

[0072] Count the harmonic mean of the precision rate and the recall rate of the data of the same type in all samples to obtain the F1 score;

[0073] According to the accuracy rate, precision rate, recall rate, and F1 score, judge whether the recognition model meets the accuracy requirements. The calculation method is as follows:

[0074] The accuracy rate represents

[0075] The precision rate represents

[0076] The recall rate represents

[0077] The F1 score represents

[0078] Among them, E TP —— The number of low-speed unruly vehicles recognized as low-speed unruly vehicles; E TN —— The number of other vehicles recognized as other vehicles; E FP—— The number of other vehicles recognized as low-speed unsociable vehicles; E FN —— The situation where low-speed unsociable vehicles are recognized as other vehicles.

[0079] The embodiment of the present application also provides a method for recognizing low-speed unsociable vehicles, which includes:

[0080] Obtain the driving behavior information of the target vehicle and the vehicles in front and behind it, and summarize it into the basic driving behavior information of the target vehicle;

[0081] Based on the basic driving behavior information of the target vehicle, calculate the speed, acceleration, and position of the vehicle relative to the vehicles in front and behind it, and summarize them with the vehicle's own driving behavior data into the characteristic indicators of the target vehicle;

[0082] Construct a membership function based on the characteristic indicators of the target vehicle, and determine whether the target vehicle belongs to a low-speed unsociable vehicle according to the total membership degree result;

[0083] Input the driving behavior basic information obtained by monitoring and the determination label obtained after calculating the characteristic indicators of the driving behavior basic information into the recognition model for training.

[0084] The embodiment of the present application also provides a method for training a low-speed unsociable vehicle recognition model, which includes:

[0085] Correspond the obtained vehicle driving behavior basic information with the vehicle label;

[0086] Construct a vehicle information dataset according to the vehicle driving behavior basic information and the vehicle label;

[0087] Based on the vehicle information dataset, train the low-speed unsociable vehicle recognition model so that the model can determine whether the target vehicle belongs to a low-speed unsociable vehicle according to the basic driving information of the target vehicle.

[0088] The embodiment of the present application also provides an electronic device, which is characterized by including a memory and a processor to implement:

[0089] Based on the vehicle trajectory video obtained by monitoring, perform vehicle detection and tracking, convert the pixel coordinates into actual coordinates, and finally export the data of each vehicle;

[0090] Calculate the characteristic values according to the exported vehicle data, construct a membership function, and determine the vehicle label;

[0091] Taking the vehicle as a unit, construct a vehicle information dataset according to the corresponding vehicle basic driving information and vehicle label;

[0092] Based on the vehicle information dataset, train the low-speed unsociable vehicle recognition model so that the model can determine whether a target vehicle belongs to a low-speed unsociable vehicle according to the basic driving information of the target vehicle.

Claims

1. A low-speed unsociable vehicle identification system, characterized in that: include: Vehicle basic information statistics module, characteristic index module, low-speed non-group determination module, and recognition model training module; The vehicle basic information statistics module is used to determine the basic information and driving data of the vehicle and the front and rear vehicles based on the following processing process: analyzing the vehicles within the field of view, obtaining the driving data of the vehicle and the front and rear vehicles as the basic information of the vehicle's driving behavior in the initial stage; The characteristic index module is used to obtain basic information of vehicle driving behavior based on the following processing: calculating the information obtained in the vehicle basic information statistics module, and obtaining characteristic index data corresponding to each vehicle from speed, acceleration, speed difference, acceleration difference, and distance difference with the preceding vehicle, and speed difference, acceleration difference, and distance difference with the following vehicle; The low-speed non-grouping determination module is used to determine whether a vehicle is a low-speed non-grouping vehicle based on the following processing: Determine the threshold and weight based on the characteristic indicator data; Determine the membership function based on the threshold and weight, build a fuzzy comprehensive evaluation model, judge whether each vehicle is a low-speed unsociable vehicle, and construct a label; The identification model training module is used to train the low-speed non-gregarious vehicle identification model based on the following processing: Taking vehicles as units, constructing a vehicle information data set based on the corresponding basic information and labels of the driving behavior of the vehicles; Based on the vehicle information data set, the low-speed non-gregarious vehicle recognition model is trained so that the model determines whether the target vehicle is a low-speed non-gregarious vehicle according to the basic information of the target vehicle's driving behavior.

2. A low-speed unsociable vehicle identification system according to claim 1, characterized in that: The vehicle basic information statistics module comprehensively records the data of the vehicle and surrounding vehicles: for vehicles entering the monitoring range, both the vehicle's own indicators and its relationship with surrounding vehicles are considered.

3. A low-speed unsociable vehicle identification system according to claim 1, characterized in that: The low-speed non-group determination module is also used to perform the following steps to complete the determination of the vehicle: Counting the 5%, 25%, 75% and 95% quantiles of the characteristic index data set of the vehicle; Calculating a threshold value for the characteristic index data set of the vehicle using the interquartile range method; Calculating the weight of the characteristic index data set of the vehicle using the CRITIC weight method; The vehicle evaluation set is divided into three categories: normal vehicles, low-speed unsociable vehicles and speeding vehicles. The membership function type of each characteristic index under each evaluation set is determined respectively. According to the calculation threshold and quantile value, the membership function is constructed respectively, the membership of each characteristic index is calculated, and the total membership of the vehicle for each evaluation set is calculated according to the weight. Compare the values ​​of each total membership function and determine the evaluation set with the maximum value, which is the label of the vehicle.

4. A low-speed unsociable vehicle identification system according to claim 1, characterized in that: The recognition model training module is also used to perform the following steps to complete the transformation of input data: Standardize the basic information of vehicle driving behavior; The vehicle labels are divided into two categories: low-speed unsociable vehicles and other vehicles, and encoded respectively to form coding vectors that can be recognized by computers; The standardized basic information of vehicle driving behavior and the encoded labels are used as input data for the recognition model training module.

5. A low-speed unsociable vehicle identification system according to claim 1, characterized in that: The recognition model training module is also used to perform the following steps for training: Dividing the model input data into a training set and a test set; The low-speed unsociable vehicle recognition module is trained based on the training set to find parameters that optimize the model performance.

6. A low-speed unsociable vehicle identification system according to claim 5, characterized in that: When the recognition model training module obtains the optimal model based on the training set, it performs the following steps: The optimal parameters obtained after training the recognition model using the training set are used to identify the samples in the test set, and the model is evaluated based on the recognition results.

7. A low-speed unsociable vehicle identification system according to claim 6, characterized in that: The recognition model training module performs the following steps when judging whether the performance of the low-speed non-gregarious vehicle recognition model meets the accuracy requirement based on the model evaluation matrix: Count the accuracy of all prediction results in the low-speed non-gregarious vehicle recognition model to evaluate the overall recognition effect of the model; The precision, recall and F1 scores of the low-speed and non-gregarious vehicle samples and other vehicle samples in the low-speed and non-gregarious vehicle recognition model are respectively counted to evaluate the recognition effect of each vehicle.

8. A method for identifying low-speed unsociable vehicles, characterized in that: include: Obtain driving behavior information of the target vehicle and the vehicles ahead and behind, and summarize it into basic driving behavior information of the target vehicle; Based on the basic information of the target vehicle's driving behavior, the speed, acceleration, and position of the vehicle relative to the front and rear vehicles are calculated, and the characteristic indicators of the target vehicle are summarized with the vehicle's own driving behavior data; A membership function is constructed based on the characteristic indexes of the target vehicle, and the total membership result is used to determine whether the target vehicle is a low-speed unsociable vehicle. The basic information of driving behavior acquired through monitoring and the judgment label obtained after calculating the characteristic index of the basic information of driving behavior are input into the recognition model to train the model.

9. An electronic device, characterized in that: Includes memory and processor to implement: Based on the vehicle trajectory video obtained by monitoring, the pixel coordinates are converted into actual coordinates based on vehicle detection and tracking, and finally the data of each vehicle is exported; Calculate the characteristic value based on the exported vehicle data, construct the membership function, and determine the vehicle label; Taking vehicles as units, construct a vehicle information dataset based on the corresponding vehicle basic driving information and vehicle labels; Based on the vehicle information data set, a low-speed non-gregarious vehicle recognition model is trained so that the model determines whether the target vehicle is a low-speed non-gregarious vehicle according to basic driving information of the target vehicle.

Images