Traffic Flow Node Anomaly Identification System Based on Hierarchical Fuzzy Neural Network

Abstract

This invention discloses a traffic flow node anomaly identification system based on a hierarchical fuzzy neural network, belonging to the field of traffic management technology. The behavior recognition module identifies the behavior data of each vehicle based on a hierarchical fuzzy neural network model. The anomaly judgment module analyzes the behavior data of each vehicle based on a vehicle anomaly identification model to determine whether a vehicle exhibits abnormal behavior. When abnormal behavior is detected, an automatic adjustment module tracks the vehicle exhibiting abnormal behavior using a camera. After determining the ETC lane where the vehicle with abnormal behavior is located, it automatically reduces the recognition range of the ETC automatic settlement device at that ETC lane. This identification system can effectively identify vehicle anomalies at toll stations. Furthermore, when a vehicle with abnormal behavior is identified, it automatically adjusts the recognition range of the ETC automatic settlement device at the ETC lane where the abnormal vehicle is located, ensuring normal passage at toll stations while effectively reducing the success rate of vehicles using ETC without authorization.

Description

Technical Field

[0001] This invention relates to the field of transportation management technology, specifically to a transportation flow node anomaly identification system based on hierarchical fuzzy neural networks. Background Technology

[0002] In transportation systems, traffic flow nodes are where information converges, including intersections, toll booths, and vehicle inspection points. Effective monitoring of these nodes allows for a more comprehensive understanding of traffic conditions, leading to better traffic flow optimization and safety management. A traffic flow node anomaly identification system utilizes advanced technologies to monitor and analyze traffic flow nodes in transportation systems, enabling timely detection and identification of anomalies. Such systems can be applied to urban traffic management, intelligent transportation systems, logistics management, and other fields, contributing to improved efficiency, safety, and reliability of transportation systems.

[0003] To facilitate quick passage and payment for drivers, existing highway toll booths are equipped with ETC (Electronic Toll Collection) automatic payment devices. However, some drivers exploit loopholes in the ETC automatic payment mechanism to bypass the gate and avoid being recorded. Because the ETC automatic payment device identifies vehicles' ETC accounts based on a pre-set recognition range, it has the following drawbacks:

[0004] 1. If the preset recognition range is too small, the vehicle may need to get close enough to the ETC automatic settlement device before it can be recognized. Although this effectively reduces the success rate of unauthorized ETC use, it will reduce the recognition efficiency of the ETC automatic settlement device during normal passage, thereby reducing the efficiency of vehicle passage.

[0005] 2. If the pre-set recognition range is too large, although it can effectively improve the vehicle passage efficiency at toll stations, it will lead to an increase in the success rate of ETC fraud.

[0006] Based on this, the present invention proposes a traffic flow node anomaly identification system based on hierarchical fuzzy neural network, which can identify abnormal vehicle behavior during normal traffic flow to dynamically adjust the identification range of ETC automatic settlement device, thereby ensuring vehicle traffic efficiency while reducing the success rate of ETC fraud. Summary of the Invention

[0007] The purpose of this invention is to provide a traffic flow node anomaly identification system based on hierarchical fuzzy neural network to address the shortcomings of the prior art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a traffic flow node anomaly identification system based on a hierarchical fuzzy neural network, wherein the identification system obtains the number of ETC lanes at toll stations and the ETC automatic settlement devices used in conjunction with the ETC lanes from the toll station management system through a data acquisition module;

[0009] The image acquisition module of the recognition system acquires vehicle images through cameras set up at each ETC lane. The behavior recognition module identifies the behavior data of each vehicle based on a hierarchical fuzzy neural network model. The anomaly judgment module analyzes the behavior data of each vehicle based on a vehicle anomaly recognition model to determine whether the vehicle has abnormal behavior.

[0010] When abnormal behavior is detected in a vehicle, the automatic adjustment module tracks the vehicle with abnormal behavior through a camera. After determining the ETC lane where the vehicle with abnormal behavior is located, it automatically reduces the recognition range of the ETC automatic settlement device in that ETC lane.

[0011] The vehicle warning module tracks vehicles exhibiting abnormal behavior using cameras, generates corresponding warning signals for such vehicles, records the vehicle's license plate information, and communicates with the traffic police system to send information about vehicles exhibiting abnormal behavior and their license plate information.

[0012] In a preferred embodiment, the behavior recognition module identifies the behavior data of each vehicle based on a hierarchical fuzzy neural network model, including the following steps:

[0013] Prepare a dataset containing samples of various behaviors, label each sample with its corresponding behavior, and normalize or standardize the dataset.

[0014] Based on the preset HFNN model structure, the HFNN model structure includes an input layer, a fuzzy neural network layer, and an output layer;

[0015] Using a labeled training dataset, the weights and parameters of the HFNN model are adjusted through the backpropagation algorithm.

[0016] The performance of the HFNN model is evaluated using a validation set independent of the training set, and the parameters are adjusted based on the performance.

[0017] The extracted vehicle image data is input into the HFNN model, and the vehicle behavior data identified by the HFNN model is output through forward propagation. The output behavior data includes vehicle lane change speed, number of consecutive lane changes, and distance between the vehicle and the ETC automatic settlement device after the vehicle enters the lane.

[0018] In a preferred embodiment, the anomaly detection module establishes a vehicle anomaly identification model by including the following steps:

[0019] The vehicle behavior coefficient is obtained by comprehensively calculating the vehicle's lane-changing speed, the number of consecutive lane changes, and the distance between the vehicle entering the lane and the ETC automatic settlement device. ;

[0020] Vehicle behavior coefficient The vehicle behavior coefficient is compared with the preset first and second anomaly thresholds. The larger the threshold, the more abnormal the vehicle is. This completes the establishment of the vehicle anomaly identification model. The first anomaly threshold is less than the second anomaly threshold. The first anomaly threshold is used to determine whether the vehicle has abnormal behavior, and the second anomaly threshold is used to determine the severity of the abnormal behavior.

[0021] In a preferred embodiment, the anomaly detection module obtains real-time vehicle lane-changing speed, the number of consecutive lane changes, and the distance between vehicles entering the lane and the ETC automatic settlement device at the toll station, and then substitutes these data into the vehicle anomaly identification model to calculate the vehicle behavior coefficient. value;

[0022] If vehicle behavior coefficient If the value is less than or equal to the first abnormal threshold, it is determined that the vehicle does not exhibit abnormal behavior.

[0023] If vehicle behavior coefficient If the value is greater than the first abnormal threshold, the vehicle is judged to have abnormal behavior. If the vehicle behavior coefficient is found to be abnormal, then... If the value is ≤ the second anomaly threshold, the vehicle is judged to have a slight anomaly. If the vehicle behavior coefficient If the value is greater than the second abnormal threshold, the vehicle is judged to have a severe abnormality.

[0024] In a preferred embodiment, when a vehicle is determined to have a minor abnormality, the automatic adjustment module determines the ETC lane where the vehicle is located, obtains the original recognition range information of the current ETC lane, including the location and range parameters of the recognition device, automatically reduces the recognition range of the ETC automatic settlement device in the ETC lane where the vehicle with abnormal behavior is located, and feeds back the location information of the vehicle with abnormal behavior to relevant personnel in real time.

[0025] When a vehicle is determined to have a serious abnormality, the vehicle warning module identifies the ETC lane where the vehicle is located, generates a warning signal, saves the license plate information of the abnormal vehicle in the recording system, and automatically controls the ETC automatic settlement device of the ETC lane where the vehicle is located to close. The vehicle's license plate information and abnormal behavior information are sent to the traffic police system through network communication protocol.

[0026] In a preferred embodiment, the logic for obtaining the vehicle lane change speed is as follows: when there is a vehicle in a certain lane of the toll station, the camera in that lane is turned on and monitors the vehicles in that lane.

[0027] When the angle formed between the front of a vehicle and the lane in a certain lane is greater than a preset angle threshold, it is determined that the vehicle has changed lanes. The camera uses the Faster R-CNN object detection algorithm to detect the vehicle with lane-changing behavior and obtain the lane-changing speed of the vehicle.

[0028] In a preferred embodiment, the logic for obtaining the number of consecutive lane changes by a vehicle is as follows: when there is a vehicle in a certain lane of the toll station, the camera in that lane is turned on and monitors the vehicles in that lane.

[0029] When the angle formed between the front of a vehicle and the lane in a certain lane is greater than a preset angle threshold, it is determined that the vehicle has changed lanes.

[0030] The camera uses the Faster R-CNN object detection algorithm to identify vehicles that have changed lanes. When a vehicle changes lanes and the angle formed by the tilt of the vehicle's front and the lane is less than or equal to a preset angle threshold, the camera determines that the vehicle has changed lanes.

[0031] When a vehicle changes lanes and the angle formed by the vehicle's front tilt and the lane is greater than a preset angle threshold, the system continues to track the vehicle through the camera in the lane it is in until the vehicle stops changing lanes, and then obtains the number of consecutive lane changes by the vehicle.

[0032] In a preferred embodiment, the logic for obtaining the distance between the vehicle behind the lane and the ETC automatic settlement device is as follows: when a vehicle changes to a certain lane, the camera at that lane obtains whether there are other vehicles behind the lane-changing vehicle in that lane. When there are other vehicles, the distance between the other vehicles and the ETC automatic settlement device after the lane-changing vehicle enters the lane is estimated by the landmark lines on the lane.

[0033] In a preferred embodiment, the image acquisition module acquires vehicle images through cameras installed at each ETC lane, including the following steps:

[0034] The camera captures images of vehicles passing through the ETC lane in real time and transmits the images to the image acquisition module for processing. The captured vehicle images include continuous video streams or single-frame images.

[0035] The acquired vehicle images undergo noise reduction, brightness and contrast adjustment, and image enhancement preprocessing.

[0036] Based on object detection algorithms, vehicles passing through the channel are detected and tracked in vehicle images. Information is extracted from the detected and tracked vehicles, including vehicle position, speed, and vehicle type.

[0037] In a preferred embodiment, the image acquisition module detects and tracks vehicles passing through the channel in vehicle images based on a target detection algorithm, including vehicle position, speed, and vehicle type, comprising the following steps:

[0038] A background model is constructed using the first few frames of images. The Faster R-CNN target detection algorithm is used to detect moving vehicle targets. The Kalman filter target tracking algorithm is used to correlate the detected vehicle targets in time and form a trajectory.

[0039] For each tracked vehicle target, the vehicle target position coordinate information is extracted, the vehicle model information is identified through the vehicle model database, and the position information captured by the camera is converted into the position in the real world coordinate system using camera calibration information;

[0040] Differential operations are used to calculate the vehicle's speed information by utilizing the vehicle's position information between adjacent frames.

[0041] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0042] 1. This invention utilizes an image acquisition module in the identification system to acquire vehicle images through cameras installed at each ETC lane. A behavior recognition module identifies the behavior data of each vehicle based on a hierarchical fuzzy neural network model. An anomaly detection module analyzes the behavior data of each vehicle based on a vehicle anomaly detection model to determine if any abnormal behavior exists. When abnormal behavior is detected, an automatic adjustment module tracks the vehicle exhibiting abnormal behavior using cameras. After identifying the ETC lane where the vehicle with abnormal behavior is located, it automatically reduces the recognition range of the ETC automatic settlement device at that lane. This identification system effectively identifies vehicle anomalies at toll stations and automatically adjusts the recognition range of the ETC automatic settlement device at the ETC lane where the abnormal vehicle is located when an abnormal vehicle is detected. This ensures normal passage at toll stations while effectively reducing the success rate of vehicles using ETC without authorization.

[0043] 2. In cases where an abnormal vehicle changes lanes too quickly, changes lanes too frequently, or gets too close to the ETC automatic settlement device after changing lanes, there is a risk to the safe driving of other vehicles. Even if the recognition range of the ETC automatic settlement device is reduced, the abnormal vehicle may still be able to access the ETC device due to the close proximity of the vehicle behind. Therefore, by setting a second abnormal threshold to determine the severity of the abnormal situation, the ETC automatic settlement device can be directly shut down when a severe abnormality is detected. This not only effectively intercepts severely abnormal vehicles but also further reduces the success rate of abnormal vehicles accessing the ETC device. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0045] Figure 1 This is a system module diagram of the present invention.

[0046] Figure 2 This is a flowchart of the method of the present invention. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0048] Example: Please refer to Figure 1 and Figure 2As shown in this embodiment, the traffic flow node anomaly identification system based on hierarchical fuzzy neural network obtains the number of ETC lanes and the corresponding ETC automatic settlement devices from the toll station management system through the data acquisition module. The numbering module groups the corresponding ETC lanes and ETC automatic settlement devices into a set. Each set contains an ETC lane and the ETC automatic settlement device used in that ETC lane, and each set is numbered. The set information is sent to the automatic adjustment module and the vehicle warning module. The image acquisition module of the identification system acquires vehicle images through the camera installed at each ETC lane. The vehicle image data is sent to the behavior recognition module, which identifies the behavior data of each vehicle based on the hierarchical fuzzy neural network model. The behavioral data is sent to the anomaly judgment module. The anomaly judgment module analyzes the behavioral data of each vehicle based on the vehicle anomaly recognition model to determine whether the vehicle has abnormal behavior. The judgment result is sent to the automatic adjustment module and the vehicle warning module. When it is determined that a vehicle has abnormal behavior, the automatic adjustment module tracks the vehicle with abnormal behavior through the camera. After determining the ETC lane where the vehicle with abnormal behavior is located, it automatically reduces the recognition range of the ETC automatic settlement device at that ETC lane. The vehicle warning module tracks the vehicle with abnormal behavior through the camera, generates corresponding warning signals for the vehicle with abnormal behavior, including verbal warnings, and records the license plate information of the vehicle. The vehicle warning module also communicates with the traffic police system and sends the vehicle with abnormal behavior and its license plate information to the traffic police system.

[0049] This application utilizes an image acquisition module in its identification system to capture vehicle images through cameras installed at each ETC lane. A behavior recognition module identifies the behavior data of each vehicle based on a hierarchical fuzzy neural network model. An anomaly detection module analyzes the behavior data of each vehicle based on a vehicle anomaly detection model to determine if any abnormal behavior exists. When abnormal behavior is detected, an automatic adjustment module tracks the vehicle exhibiting abnormal behavior using cameras. Once the ETC lane containing the abnormal vehicle is identified, the system automatically reduces the recognition range of the ETC automatic settlement device at that lane. This identification system effectively identifies vehicle anomalies at toll stations and automatically adjusts the recognition range of the ETC automatic settlement device in the lane where the abnormal vehicle is located when an abnormal behavior is detected. This ensures normal passage at toll stations while effectively reducing the success rate of vehicles using ETC without authorization.

[0050] The identification system obtains the number of ETC lanes at the toll station and the ETC automatic settlement devices used with the ETC lanes from the toll station management system through the data acquisition module. Specifically, it obtains relevant information by connecting to the toll station management system or data source, and uses appropriate protocols and interfaces to send a request to the toll station management system to obtain the number of ETC lanes at the toll station and the information on the ETC automatic settlement devices used with the lanes. This determines the number of ETC lanes at the toll station for subsequent matching and identification. The system obtains the number of ETC lanes from the toll station management system, which typically includes the unique identifier and location information of each lane.

[0051] The system acquires information about the ETC automatic settlement device used with each ETC lane. For each ETC lane, it retrieves information about the ETC automatic settlement device used with it from the toll station management system, which may include device ID, model, status, etc. The data obtained from the toll station management system is processed and stored to facilitate subsequent identification. The system processes the collected ETC lane number and corresponding automatic settlement device information and stores it in the system database or other data storage media. The system updates the number of ETC lanes and the matching automatic settlement device information to ensure that the system has the latest data. The system updates the number of ETC lanes and the matching automatic settlement device information from the toll station management system periodically or as needed to keep the system synchronized with the actual situation.

[0052] Handle potential anomalies, such as inability to connect to the toll station management system or data acquisition errors, and implement appropriate anomaly handling mechanisms, such as logging, issuing alarms, or taking other measures, to ensure the stability and reliability of the system. Send notifications to relevant personnel or the system to report changes in the number of ETC lanes and automatic settlement device information. After successfully acquiring and updating data, the system can send notifications to relevant personnel through a notification mechanism to keep them informed of toll station configuration changes in a timely manner.

[0053] The numbering module groups the corresponding ETC lanes and ETC automatic settlement devices into a set. Each set contains an ETC lane and the ETC automatic settlement device used in that ETC lane, and each set is assigned a number, as follows:

[0054] Each ETC lane and its corresponding ETC automatic settlement device are divided into a set, and a unique number is generated for each set. Each ETC lane is traversed and its matching ETC automatic settlement device is combined into a set. Each set is assigned a unique number, which can be an auto-incrementing number or other unique identifier. The information of each set is stored, including the number, the included ETC lanes, and the ETC automatic settlement device information.

[0055] The information of each set is stored in the system database or other data storage media to ensure that it can be queried and updated at any time. In response to changes in ETC lanes and automatic settlement devices at toll stations, the information of the sets is updated and maintained in a timely manner. The information of the sets is updated periodically or when the system receives a change notification, including the addition of new ETC lanes, deletion of ETC lanes, or device changes. A query interface is provided to find the corresponding ETC lane and the ETC automatic settlement device used by the set number. The query algorithm is implemented so that users can quickly retrieve the corresponding ETC lane and automatic settlement device information by entering the set number.

[0056] Handle potential anomalies, such as querying non-existent sets or number conflicts, and implement anomaly handling mechanisms, such as returning error messages or logging, to track and troubleshoot problems. When set information changes, send notifications to relevant personnel or the system to ensure that the system is synchronized with the actual configuration. After successfully updating set information, send notifications to relevant personnel through a notification mechanism to keep abreast of changes in the configuration of ETC lanes and automatic settlement devices at toll stations.

[0057] The image acquisition module of the recognition system acquires vehicle images through cameras installed at each ETC lane, specifically:

[0058] The system acquires real-time images of vehicles passing through the ETC lane. Cameras periodically or in real-time capture images of vehicles in the lane and transmit them to the system for processing. The acquired images can be continuous video streams or single-frame images. The acquired vehicle images are preprocessed to improve the accuracy of subsequent analysis. Preprocessing operations such as denoising, adjusting brightness and contrast, and image enhancement are performed to ensure that the image quality meets the analysis requirements. Based on target detection algorithms, vehicles passing through the lane are detected and tracked in the images. The system accurately detects and tracks vehicles in the images and extracts key information from the detected and tracked vehicles, such as vehicle position, speed, and vehicle type. Information extraction is performed on the detected and tracked vehicles to form vehicle image data for subsequent analysis and recognition.

[0059] Based on object detection algorithms, vehicles passing through a channel are detected and tracked in vehicle images. Information is extracted from the detected and tracked vehicles, including vehicle position, speed, and vehicle type. The process includes the following steps:

[0060] Camera calibration is performed to correct lens distortion and ensure that the vehicle's position in the image matches its actual position. A background model is constructed using the first few frames of images so that changes in subsequent frames can be identified as moving objects. The Faster R-CNN object detection algorithm is used to detect moving vehicle targets. The Kalman filter object tracking algorithm is used to correlate the detected targets in time to form a trajectory. For each tracked target, its position coordinates, speed, and other information are extracted. Vehicle type information can be identified using image recognition technology or a vehicle type database. Using camera calibration information, the position information captured by the camera is converted into a position in the real world coordinate system to obtain accurate vehicle position. Differential operations are used to calculate the vehicle's speed using the vehicle position information between adjacent frames. Image recognition technology or a combination of vehicle type databases is used to identify the vehicle and obtain vehicle type information. The vehicle information, including position, speed, and vehicle type, is output to ensure data queryability and long-term storage.

[0061] The process involves constructing a background model using the first few frames of imagery, detecting moving vehicle targets using the Faster R-CNN object detection algorithm, and then using a Kalman filter object tracking algorithm to correlate the detected vehicle targets over time and form their trajectories. This includes the following steps:

[0062] A background model of the scene is constructed to distinguish moving targets in the scene. The static background of the scene is modeled using the first few frames of images, usually using statistical methods or deep learning-based methods. The background model is updated periodically to adapt to possible changes in the scene. Faster R-CNN or other object detection algorithms are used to detect objects in each frame of the image, identify vehicle targets, and obtain the feature vector of each detected target for tracking.

[0063] A Kalman filter is initialized for each detected target, along with its state and covariance matrix. In each frame, the Kalman filter is used for state prediction to estimate the target's position in the current frame. The state of the Kalman filter is updated using the target detection results to correct the predicted position of the target. The target position estimated by the Kalman filter is used to form the trajectory of the moving target. The detection results in the current frame are correlated with the previous trajectories to ensure that each target is properly updated on the trajectory. The trajectory of each target is stored and managed, including information such as position and velocity. Trajectories can be filtered to remove unstable or erroneous trajectories.

[0064] For each tracked vehicle target, the vehicle target's position coordinate information is extracted. The vehicle model information is identified through a vehicle model database. Using camera calibration information, the position information captured by the camera is converted into a position in the real-world coordinate system, including the following steps:

[0065] The current position coordinates of the vehicle target are obtained from the Kalman filter algorithm or other target tracking algorithms. The vehicle target is identified through the vehicle model database to obtain vehicle type information. Features such as vehicle shape and color are extracted from each tracked vehicle target. The extracted features are matched with the information in the vehicle model database to identify the vehicle type. Camera calibration information, including the camera's intrinsic and extrinsic parameters, is obtained. The camera's intrinsic parameters, such as focal length and principal point coordinates, are obtained using methods such as camera calibration boards. The camera's extrinsic parameters, i.e., the camera's position and orientation in the world coordinate system, are calibrated. The position information is then converted into the position in the actual world coordinate system.

[0066] The camera calibration information is used to convert the position information captured by the camera into the position in the real world coordinate system. The camera coordinate system to world coordinate system transformation is carried out by using the camera's intrinsic and extrinsic parameters to transform the position information captured by the camera from the camera coordinate system to the world coordinate system. According to the scale of the actual scene, possible scale transformations are performed to ensure that the position information is consistent with the real world coordinate system. The transformed real world coordinate information is saved or applied to the system. The real world coordinate information of each vehicle target is saved to the database or log for subsequent analysis or application. The real world coordinate information is applied to other systems, such as traffic monitoring, route planning, etc.

[0067] Using differential operations to calculate vehicle speed information based on vehicle position information between adjacent frames includes the following steps:

[0068] The vehicle position information between adjacent frames is obtained from the vehicle target tracking system. For each frame, the position coordinate information of the vehicle target is obtained from the tracking algorithm. The change in position is obtained by performing a difference operation on the vehicle position information between adjacent frames. For each vehicle target, the difference (displacement) of its position between two adjacent frames is calculated, that is, the position of the current frame minus the position of the previous frame.

[0069] The vehicle's speed is calculated using the position change obtained through differential analysis. Using the differential information of the vehicle's position and the time interval between frames, the average speed of the vehicle is calculated using the following formula: Speed ​​= Position Change / Time Interval. In practice, the speed in the horizontal and vertical directions can be calculated separately, or the magnitude of the speed can be calculated. Further processing of the obtained speed data is possible, such as smoothing to remove noise or outliers, or other processing operations as needed. The obtained speed information can be saved or applied to the system, either by storing it in a database or log for subsequent analysis or application.

[0070] The behavior recognition module identifies the behavior data of each vehicle based on a hierarchical fuzzy neural network model, specifically:

[0071] Prepare a dataset containing various behavioral samples, label each sample with its corresponding behavior, such as normal driving, parking, lane changing, and emergency braking. Normalize or standardize the feature data to prevent certain features from having an excessive impact on the training of the HFNN model. Design the structure of the HFNN model, including the input layer, fuzzy neural network layer, and output layer, and determine the network connections and weights. Using the labeled training dataset, adjust the weights and parameters of the HFNN model through training methods such as backpropagation to optimize the performance of the HFNN model. Evaluate the model's performance using a validation set independent of the training set, and adjust the parameters based on the performance to avoid overfitting or underfitting. Input the extracted feature data into the HFNN model, and obtain the output of the HFNN model, i.e., the vehicle's behavioral data, through forward propagation. The output behavioral data includes the vehicle's lane changing speed, the number of consecutive lane changes, and the distance between the vehicle and the ETC automatic settlement device after the vehicle enters the lane.

[0072] Preparing a dataset containing samples of various behaviors, labeling each sample with its corresponding behavior, and normalizing or standardizing the dataset includes the following steps:

[0073] Vehicle behavior data is acquired using sensors, cameras, and other devices to ensure the dataset covers various real-world scenarios. Each data sample is labeled manually or automatically to indicate whether it represents normal behavior or an abnormal behavior, such as parking, acceleration, or lane changing. Preprocessing operations include noise removal, handling missing values, and data sampling.

[0074] Based on the task requirements, select an appropriate feature extraction method, such as extracting features like position, speed, and acceleration from vehicle images. Adjust the scale of the features to prevent certain features from having an excessive impact on model training. Divide the dataset into training and test sets for model training and evaluation. Randomly or according to certain rules, divide the data to ensure that the sample distribution of the training and test sets is uniform. Save the processed dataset as a file for model training and testing. Save the processed dataset as a formatted file, such as CSV or JSON, for loading and use in subsequent tasks.

[0075] Based on the pre-defined HFNN model structure, which includes an input layer, a fuzzy neural network layer, and an output layer, the following steps are included:

[0076] The number of nodes in the input layer is determined based on the number of features. For example, if each sample has N features, then the number of nodes in the input layer is N. A fuzzy set is designed for each node in the input layer, for example, using a triangle or trapezoidal function. The fuzzy rules are determined, i.e., how to combine the fuzzy sets of the input to generate the output of the fuzzy neural network layer. The number of nodes in the fuzzy neural network layer is determined based on the task requirements and the complexity of the model.

[0077] Construct the output layer of the HFNN model, define the output format and possible categories, and determine the number of nodes in the output layer based on the number of categories in the task. For example, a binary classification task may have one output node, while a multi-class classification task may have multiple output nodes. Select the activation function for the output layer. For example, the sigmoid activation function can be used for binary classification tasks, while the softmax activation function can be used for multi-class classification tasks. Connect the input layer, the fuzzy neural network layer, and the output layer, and initialize the connection weights. Random initialization or other suitable weight initialization methods can be used. Define the loss function, optimizer, and evaluation metric to prepare for model training. Compile the HFNN model using an appropriate loss function, optimizer, and evaluation metric.

[0078] Using a labeled training dataset, the HFNN model is trained via backpropagation, and the weights and parameters are adjusted through the following steps:

[0079] Load the labeled training dataset, ensuring each sample has a corresponding label. Load the training dataset into the model, including the input data and corresponding labels. Perform a forward propagation through the model, calculating the model's output. Pass the input data to the model and calculate the output. Calculate the difference between the model's output and the actual labels, using a loss function (e.g., cross-entropy loss) to calculate the difference. Calculate the gradient of the loss function with respect to the model parameters to prepare for the optimization process. Finally, use backpropagation to calculate the gradient of the loss function with respect to the model parameters, which typically involves the application of the chain rule.

[0080] Based on the calculated gradient, update the model's weights and parameters. Use optimization algorithms (such as gradient descent, Adam, etc.) to repeatedly execute the forward propagation, loss calculation, backpropagation, and parameter update process based on the gradient. Pass the entire training dataset to the model multiple times to continuously adjust the model parameters and improve model performance. Use a validation set independent of the training set to evaluate the model's performance on unseen data. Based on the model's performance on the validation set, adjust hyperparameters to optimize the model. Based on the validation set results, adjust hyperparameters (such as learning rate, model complexity, etc.) to improve the model's generalization ability. Save the trained model for future use. Save the model to a file for loading and use in other tasks.

[0081] Evaluating the performance of the HFNN model using a validation set independent of the training set, and tuning parameters based on performance, includes the following steps:

[0082] A validation set is created from data independent of the training set. A portion of the data is retained as the validation set to ensure that the sample distribution of the validation set is similar to that of the training set. The validation set is used to evaluate the performance of the HFNN model on unseen data. The validation set is then input into the trained model, and the model's performance metrics on the validation set, such as accuracy, precision, and recall, are calculated. The model's performance on the validation set is analyzed, and the model's performance metrics on the validation set are observed to understand the model's accuracy and other relevant metrics.

[0083] Adjust the hyperparameters of the HFNN model based on performance. Consider adjusting hyperparameters such as learning rate, model complexity, and regularization to improve the model's generalization performance. After adjusting the parameters, retrain the HFNN model using the entire training set (including the validation set) to ensure that the model can fully learn the new parameter configuration. Continuously adjust the parameters and evaluate the performance in multiple iterations. If necessary, perform parameter adjustments and retraining multiple times until satisfactory performance is obtained. Save the trained model for future use. After completing the parameter adjustment, save the final model for use in practical applications.

[0084] The extracted vehicle image data is input into the HFNN model, and the vehicle behavior data identified by the HFNN model is output through forward propagation. The output behavior data includes vehicle lane change speed, number of consecutive lane changes, and distance between the vehicle and the ETC automatic settlement device after the vehicle enters the lane. The steps include:

[0085] Acquire vehicle image data captured by the camera, extract vehicle image data from the images or videos captured by the camera, ensure the quality and integrity of the data, preprocess the vehicle image data to meet the input requirements of the HFNN model, and perform scaling, normalization or other necessary preprocessing operations on the image data to ensure that the data format input into the HFNN model is correct.

[0086] The preprocessed vehicle image data is input into the HFNN model for forward propagation. The forward propagation method of the HFNN model is called to pass the preprocessed vehicle image data to the model and obtain the output results of the model. The output of the HFNN model is analyzed to obtain the recognized vehicle behavior data. Based on the structure of the model output, data containing information such as vehicle lane change speed, number of consecutive lane changes, and distance of vehicles behind the vehicle entering the lane from the ETC automatic settlement device are extracted.

[0087] Analyze the obtained behavioral data to understand the vehicle's behavior in the channel. Based on business needs and task objectives, further process and analyze the output behavioral data. Based on the analysis results of the behavioral data, take corresponding actions or make records. Depending on the identified vehicle behavior, it may be necessary to trigger alarms, log entries, or use the results in other systems.

[0088] The logic for obtaining vehicle lane change speed is as follows: When there is a vehicle in a toll station lane (including manual lanes and ETC lanes), the camera in that lane is turned on and monitors the vehicles in that lane. When the angle formed by the tilt of the front of a vehicle in a lane and the lane is greater than a preset angle threshold (the angle threshold can be 30 degrees or 45 degrees, which is set by different toll stations), it is determined that the vehicle has changed lanes. The camera uses the Faster R-CNN object detection algorithm to detect vehicles with lane change behavior and obtains the vehicle's lane change speed. Generally speaking, if a vehicle is changing lanes normally, it will change lanes at a slower speed to ensure safe lane changing. The faster the vehicle changes lanes, the more likely the vehicle is to use ETC without authorization.

[0089] The logic for obtaining the number of consecutive lane changes by a vehicle is as follows: When there is a vehicle in a certain lane of the toll station (including manual lanes and ETC lanes), the camera in that lane is turned on and monitors the vehicles in that lane. When the angle formed by the tilt of the front of a vehicle in a certain lane with the lane is greater than a preset angle threshold (the angle threshold can be 30 degrees or 45 degrees, which is set by different toll stations), it is determined that the vehicle has changed lanes. The camera uses the Faster R-CNN object detection algorithm to detect vehicles that have changed lanes. When the vehicle changes lanes and the angle formed by the tilt of the front of the vehicle with the lane is less than or equal to the preset angle threshold, it is determined that the vehicle has changed lanes. When the vehicle changes lanes and the angle formed by the tilt of the front of the vehicle with the lane is still greater than the preset angle threshold, the vehicle continues to be tracked by other cameras (the camera in the lane where the vehicle is located is used for tracking) until the vehicle stops changing lanes, and then the number of consecutive lane changes by the vehicle is obtained.

[0090] Generally speaking, vehicles are not allowed to change lanes continuously. Continuous lane changes pose a significant safety risk to traffic safety. Therefore, the more times a vehicle changes lanes continuously, the more likely it is to be using an ETC (Electronic Toll Collection) fraudulently.

[0091] The logic for obtaining the distance between vehicles behind a vehicle entering a lane and the ETC automatic settlement device is as follows: When using ETC, it is necessary to use the ETC device of another vehicle to complete the process. Therefore, when a vehicle changes to a certain lane, the camera at that lane obtains whether there are other vehicles behind the vehicle that changed lanes. If there are other vehicles, the distance between the other vehicles and the ETC automatic settlement device after the vehicle that changed lanes enters the lane is estimated by using the landmark lines on the lane.

[0092] Generally speaking, drivers who use ETC (Electronic Toll Collection) to cheat usually use their rearview mirrors to observe which lane is about to be entered, and then quickly change lanes to the front of the vehicle behind it before it enters the lane, so as to achieve the purpose of cheating on ETC. Therefore, when there are vehicles changing lanes in the lane, the smaller the distance between the vehicle behind that vehicle and the ETC automatic settlement device, the more likely there is to be cheating on ETC in that lane.

[0093] The anomaly detection module analyzes the behavioral data of each vehicle based on a vehicle anomaly identification model to determine whether the vehicle exhibits abnormal behavior. Specifically:

[0094] The establishment of a vehicle anomaly detection model includes the following steps:

[0095] The vehicle behavior coefficient is obtained by comprehensively calculating the vehicle's lane-changing speed, the number of consecutive lane changes, and the distance between the vehicle entering the lane and the ETC automatic settlement device. The calculation expression is:

[0096] In the formula, For vehicle lane changing speed, This refers to the number of consecutive lane changes by a vehicle. This refers to the distance between the vehicle and the ETC automatic settlement device after the vehicle enters the lane. , These are the proportional coefficients representing the distance between the vehicle and the ETC automatic settlement device after the vehicle enters the lane, and the vehicle's lane-changing speed. , All are greater than 0, proportionality coefficient , This is used to standardize the distance between vehicles and the ETC automatic settlement device after a vehicle enters the lane, as well as the vehicle's lane-changing speed, to avoid the excessive impact of the distance between vehicles and the ETC automatic settlement device or the vehicle's lane-changing speed on the vehicle anomaly identification model.

[0097] Vehicle behavior coefficient The vehicle behavior coefficient is compared with the preset first and second anomaly thresholds. The larger the threshold, the more abnormal the vehicle is, thus completing the establishment of the vehicle anomaly identification model. The first anomaly threshold is less than the second anomaly threshold. The first anomaly threshold is used to determine whether the vehicle has abnormal behavior, and the second anomaly threshold is used to determine the severity of the abnormal behavior.

[0098] The anomaly detection module obtains real-time data on vehicle lane-changing speeds, the number of consecutive lane changes, and the distance between vehicles behind the tollbooth and the ETC automatic settlement device. This data is then input into the vehicle anomaly identification model to calculate vehicle behavior coefficients. value;

[0099] If vehicle behavior coefficient If the value is less than or equal to the first abnormal threshold, it is determined that the vehicle does not exhibit abnormal behavior.

[0100] If vehicle behavior coefficient If the value is greater than the first anomaly threshold, the vehicle is judged to have abnormal behavior. Specifically, the first anomaly threshold is greater than the vehicle behavior coefficient. If the value is ≤ the second anomaly threshold, the vehicle is judged to have a slight anomaly. If the vehicle behavior coefficient If the value is greater than the second abnormal threshold, the vehicle is judged to have a severe abnormality.

[0101] When abnormal vehicle behavior is detected, the automatic adjustment module tracks the vehicle using a camera. After determining the ETC lane where the abnormal vehicle is located, it automatically reduces the recognition range of the ETC automatic settlement device in that lane. Specifically:

[0102] When a vehicle is found to have a minor abnormality, the system determines the ETC lane in which the vehicle is located, obtains the original recognition range information of the current ETC lane, including the location and range parameters of the recognition device, and automatically reduces the recognition range of the ETC automatic settlement device in the lane where the abnormal vehicle is located. The system also feeds back the location information of the abnormal vehicle to relevant personnel or systems in real time so that they can take further measures, such as notifying traffic police or issuing an alarm.

[0103] The vehicle warning module tracks vehicles exhibiting abnormal behavior using cameras, generates corresponding warning signals for such vehicles (including verbal warnings), and records the vehicle's license plate information. Furthermore, the vehicle warning module communicates with the traffic police system, sending information about vehicles exhibiting abnormal behavior and their license plate details to the traffic police system. Specifically:

[0104] When a vehicle is determined to have a serious abnormality, the system identifies the ETC lane in which the vehicle is located and generates a warning signal. For example, a warning may be issued to the driver via a loudspeaker system. Alternatively, vehicle lights or screen displays may be used to issue a warning. The system records the license plate information of the vehicle exhibiting abnormal behavior, as well as other key information such as time and location. The vehicle warning module automatically controls the ETC automatic settlement device in the ETC lane in which the vehicle is located to close. The system then transmits the vehicle's license plate information and abnormal behavior information to the traffic police system via a network communication protocol to ensure timely information transmission.

[0105] When an abnormal vehicle changes lanes too quickly, changes lanes too many times consecutively, or gets too close to the ETC automatic settlement device after changing lanes, it not only poses a risk to the safe driving of other vehicles, but also, even if the recognition range of the ETC automatic settlement device is reduced, the abnormal vehicle may still be able to access the ETC due to the close proximity of the vehicle behind. Therefore, by setting a second abnormal threshold to determine the severity of the abnormal situation of the abnormal vehicle, the ETC automatic settlement device can be directly shut down when a severe abnormality is detected. This not only effectively intercepts severely abnormal vehicles, but also further reduces the success rate of abnormal vehicles accessing the ETC.

[0106] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0107] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0108] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A traffic flow node anomaly identification system based on hierarchical fuzzy neural network, characterized in that: The identification system obtains the number of ETC lanes at the toll station and the ETC automatic settlement devices used with the ETC lanes from the toll station management system through the data acquisition module; the numbering module assigns the corresponding ETC lanes and ETC automatic settlement devices to a set, each set containing the ETC lane and the ETC automatic settlement device used in that ETC lane, and each set has a number. The set information is sent to the automatic adjustment module and the vehicle warning module. The image acquisition module of the recognition system acquires vehicle images through cameras installed at each ETC lane. The behavior recognition module identifies the behavior data of each vehicle based on a hierarchical fuzzy neural network model. The anomaly detection module analyzes the behavior data of each vehicle based on a vehicle anomaly detection model to determine whether the vehicle exhibits abnormal behavior. The output behavior data includes vehicle lane-changing speed, the number of consecutive lane changes, and the distance between the vehicle and the ETC automatic settlement device after the vehicle enters the lane. The anomaly detection module establishes the vehicle anomaly detection model through the following steps: The vehicle behavior coefficient is obtained by comprehensively calculating the vehicle's lane-changing speed, the number of consecutive lane changes, and the distance between the vehicle entering the lane and the ETC automatic settlement device. ; Vehicle behavior coefficient The vehicle behavior coefficient is compared with the preset first and second anomaly thresholds. The larger the threshold, the more abnormal the vehicle is, thus establishing a vehicle anomaly identification model. The first anomaly threshold is less than the second anomaly threshold. The first threshold determines whether the vehicle exhibits abnormal behavior, while the second threshold determines the severity of the abnormal behavior. The anomaly detection module obtains real-time data from the toll station, including vehicle lane-changing speed, the number of consecutive lane changes, and the distance between vehicles entering the lane and the ETC automatic settlement device. These data are then used to calculate the vehicle behavior coefficient using the vehicle anomaly identification model. value; If vehicle behavior coefficient If the value is less than or equal to the first abnormal threshold, it is determined that the vehicle does not exhibit abnormal behavior. If vehicle behavior coefficient If the value is greater than the first abnormal threshold, the vehicle is judged to have abnormal behavior. If the vehicle behavior coefficient is found to be abnormal, then... If the value is ≤ the second anomaly threshold, the vehicle is judged to have a slight anomaly. If the vehicle behavior coefficient If the value is greater than the second abnormal threshold, the vehicle is judged to have a severe abnormality; When abnormal vehicle behavior is detected, the automatic adjustment module tracks the vehicle with abnormal behavior using a camera. After determining the ETC lane where the vehicle with abnormal behavior is located, it automatically reduces the recognition range of the ETC automatic settlement device in that ETC lane. Specifically, when a minor abnormality is detected, the automatic adjustment module determines the ETC lane where the vehicle is located, obtains the original recognition range information of the current ETC lane, including the location and range parameters of the recognition device, and automatically reduces the recognition range of the ETC automatic settlement device in the ETC lane where the vehicle with abnormal behavior is located, and feeds back the location information of the vehicle with abnormal behavior to relevant personnel in real time. When a vehicle is determined to have a serious abnormality, the vehicle warning module determines the ETC lane where the vehicle is located, generates a warning signal, saves the license plate information of the abnormal vehicle in the recording system, and automatically controls the ETC automatic settlement device of the ETC lane where the vehicle is located to close. The vehicle's license plate information and abnormal behavior information are sent to the traffic police system through the network communication protocol. The vehicle warning module tracks vehicles exhibiting abnormal behavior using cameras, generates corresponding warning signals for such vehicles, records the vehicle's license plate information, and communicates with the traffic police system to send information about vehicles exhibiting abnormal behavior and their license plate information.

2. The traffic flow node anomaly identification system based on hierarchical fuzzy neural network according to claim 1, characterized in that: The behavior recognition module identifies the behavior data of each vehicle based on a hierarchical fuzzy neural network model, including the following steps: Prepare a dataset containing samples of various behaviors, label each sample with its corresponding behavior, and normalize or standardize the dataset. Based on the preset HFNN model structure, the HFNN model structure includes an input layer, a fuzzy neural network layer, and an output layer; Using a labeled training dataset, the weights and parameters of the HFNN model are adjusted through the backpropagation algorithm. The performance of the HFNN model is evaluated using a validation set independent of the training set, and the parameters are adjusted based on the performance. The extracted vehicle image data is input into the HFNN model, and the vehicle behavior data identified by the HFNN model is output through forward propagation.

3. The traffic flow node anomaly identification system based on hierarchical fuzzy neural network according to claim 2, characterized in that: The logic for obtaining the vehicle lane change speed is as follows: when there is a vehicle in a certain lane of the toll station, the camera in that lane is turned on and monitors the vehicles in that lane. When the angle formed between the front of a vehicle and the lane in a certain lane is greater than a preset angle threshold, it is determined that the vehicle has changed lanes. The camera uses the Faster R-CNN object detection algorithm to detect the vehicle with lane-changing behavior and obtain the lane-changing speed of the vehicle.

4. The traffic flow node anomaly identification system based on hierarchical fuzzy neural network according to claim 2, characterized in that: The logic for obtaining the number of consecutive lane changes by a vehicle is as follows: when there is a vehicle in a certain lane of the toll station, the camera in that lane is turned on and monitors the vehicles in that lane. When the angle formed between the front of a vehicle and the lane in a certain lane is greater than a preset angle threshold, it is determined that the vehicle has changed lanes. The camera uses the Faster R-CNN object detection algorithm to identify vehicles that have changed lanes. When a vehicle changes lanes and the angle formed by the tilt of the vehicle's front and the lane is less than or equal to a preset angle threshold, the camera determines that the vehicle has changed lanes. When a vehicle changes lanes and the angle formed by the vehicle's front tilt and the lane is greater than a preset angle threshold, the system continues to track the vehicle through the camera in the lane it is in until the vehicle stops changing lanes, and then obtains the number of consecutive lane changes by the vehicle.

5. The traffic flow node anomaly identification system based on hierarchical fuzzy neural network according to claim 2, characterized in that: The logic for obtaining the distance between vehicles behind the vehicle entering the lane and the ETC automatic settlement device is as follows: When a vehicle changes to a certain lane, the camera at that lane obtains whether there are other vehicles behind the vehicle changing lanes in that lane. If there are other vehicles, the distance between the other vehicles and the ETC automatic settlement device after the vehicle changing lanes enters the lane is estimated by the landmark lines on the lane.

6. The traffic flow node anomaly identification system based on hierarchical fuzzy neural network according to claim 5, characterized in that: The image acquisition module acquires vehicle images through cameras installed at each ETC lane, including the following steps: The camera captures images of vehicles passing through the ETC lane in real time and transmits the images to the image acquisition module for processing. The captured vehicle images include continuous video streams or single-frame images. The acquired vehicle images undergo noise reduction, brightness and contrast adjustment, and image enhancement preprocessing. Based on object detection algorithms, vehicles passing through the channel are detected and tracked in vehicle images. Information is extracted from the detected and tracked vehicles, including vehicle position, speed, and vehicle type.

7. The traffic flow node anomaly identification system based on hierarchical fuzzy neural network according to claim 6, characterized in that: The image acquisition module detects and tracks vehicles passing through the channel in vehicle images based on a target detection algorithm, including vehicle position, speed, and vehicle type, and includes the following steps: A background model is constructed using the first few frames of images. The Faster R-CNN target detection algorithm is used to detect moving vehicle targets. The Kalman filter target tracking algorithm is used to correlate the detected vehicle targets in time and form a trajectory. For each tracked vehicle target, the vehicle target position coordinate information is extracted, the vehicle model information is identified through the vehicle model database, and the position information captured by the camera is converted into the position in the real world coordinate system using camera calibration information; Differential operations are used to calculate the vehicle's speed information by utilizing the vehicle's position information between adjacent frames.

Images