A method for predicting traffic congestion caused by abnormal driving of large vehicles

Abstract

The present invention relates to the field of Internet security services for traffic congestion, and specifically to a method for predicting traffic congestion caused by abnormal driving of large vehicles, comprising: utilizing cameras and radar equipment to collect real-time image information and point cloud information of road traffic; processing the image information and point cloud information to obtain target information of each target, then subjecting the target to feature-level fusion using the cameras and radar equipment, and tracking and identifying the target; based on the road traffic information, performing positive prediction and detection of abnormal driving behavior of large vehicles, including large vehicles illegally entering fast lanes, resulting in overall speed restrictions in fast lanes, and large vehicles breaking down on main roads, causing congestion; when abnormal behavior of large vehicles is detected, traffic information features are input into a forward prediction model trained by deep learning for calculation to obtain a traffic congestion prediction result. The present invention provides effective reduction of traffic problems as soon as possible after congestion occurs.

Description

Technical Field

[0001] The present invention relates to the field of Internet security services for traffic congestion, and in particular to a method for predicting traffic congestion caused by abnormal driving of large vehicles. Background Art

[0002] With the continuous development of urban transportation, traffic congestion caused by erratic driving of large vehicles has become increasingly prominent. The continuous expansion of road networks has led to an increase in the number and complexity of road networks, and erratic driving of large vehicles has become a major cause of road congestion. Large highway networks face increasingly severe traffic management challenges. Traditional traffic congestion prediction methods have limitations in addressing erratic driving of large vehicles, necessitating the need for more advanced deep learning-based technologies to improve accuracy and real-time performance.

[0003] Current traffic congestion prediction technologies face challenges in identifying abnormal driving behavior by large vehicles. The complex behavior of drivers and the difficulty in quantitatively analyzing these behaviors make it difficult for existing methods to accurately predict traffic congestion caused by abnormal driving behavior by large vehicles. Key shortcomings include:

[0004] (1) Lack of real-time performance:

[0005] Traditional methods take a long time to preprocess data and extract features, hindering the entire real-time prediction process. These methods require real-time traffic data to be transmitted over the network to a central server or cloud for processing. Network transmission latency can affect real-time performance. The sheer scale of traffic data required by traditional methods can be time-consuming. Processing multiple types of information requires even more time, and model construction can be complex and cumbersome.

[0006] (2) Abnormal driving behavior of large vehicles is difficult to judge:

[0007] Diverse abnormal behaviors: Abnormal driving behaviors of large vehicles cover a wide range of situations, such as illegally entering lanes, breaking down, and emergency braking. This diversity complicates the design of a universal and accurate judgment model, as different types of abnormal behaviors may require different features and algorithms.

[0008] (3) Insufficient prediction accuracy:

[0009] Traffic conditions on the road are influenced by numerous factors, including weather, road conditions, and changes in traffic regulations. These complex environmental changes make it difficult for prediction models to adapt to these diverse factors, resulting in insufficient prediction accuracy. Furthermore, traditional methods rely too heavily on human experience at the decision-making end, lacking standardized criteria and objective quantification and standardization. Summary of the Invention

[0010] The present invention aims to provide a method for predicting traffic congestion caused by abnormal driving of large vehicles, so as to solve the problems of traditional methods for traffic congestion, such as insufficient real-time performance, difficulty in judging abnormal driving behavior of large vehicles and insufficient prediction accuracy.

[0011] The method for predicting traffic congestion caused by abnormal driving of large vehicles in this solution includes the following steps:

[0012] Step 1: Use cameras and radar equipment to collect road traffic information in real time. Road traffic information includes image information and point cloud information;

[0013] Step 2: Process the image and point cloud information to obtain target information for each target. This information is then fused at the feature level using the camera and radar equipment to track and identify the target. This target information includes vehicle location, type, speed, heading angle, and lane-level information. It also includes traffic flow and driving violations, including illegal lane changes, illegal parking, and driving against traffic.

[0014] Step 3: Based on the road traffic information in step 1, positive prediction is performed to detect abnormal driving behaviors of large vehicles. Abnormal driving behaviors of large vehicles include large vehicles illegally entering the fast lane, resulting in overall speed restrictions in the fast lane, and large vehicles breaking down on the main road, causing congestion.

[0015] Step 4: When abnormal behavior of a large vehicle is detected, the traffic information features are input into the forward prediction model trained by deep learning for calculation to obtain the traffic congestion prediction result.

[0016] The beneficial effects of this program are:

[0017] By collecting data through cameras and radars, we can complement each other's disadvantages, ensure the accuracy and precision of data collection, and fuse the two types of data at the feature level to obtain comprehensive traffic information. Automatic forward prediction of traffic conditions can then be performed to detect and respond to abnormal driving conditions of large vehicles early, thereby providing effective data support for early prediction of traffic congestion and providing effective measures to reduce the occurrence of traffic problems as soon as possible after congestion occurs.

[0018] Furthermore, in the step 2, pixel coordinate information and target categories of the target in the image information are identified by using the YOLOv5 model in computer vision, and the target categories include non-motor vehicles, motorcycles, cars, buses, and trucks;

[0019] Preprocess the point cloud information, then perform registration, integrate point cloud data from multiple locations or times, extract features to obtain edge and plane information in the point cloud, perform object segmentation and recognition, divide the point cloud into different objects, and classify the target categories.

[0020] The beneficial effect is: through data processing, useful information can be extracted from the lidar point cloud, providing key data support for road spatial perception and environmental understanding, and providing accurate data support for subsequent data fusion.

[0021] Furthermore, in step 2, the targets are tracked by using a Deepsort model to identify the motion trajectory of each target;

[0022] The identified trajectory is mapped to the world coordinate system through the Homography matrix to show the real motion trajectory of the target on the map;

[0023] The Deepsort model is used to compare the positions of the targets in the previous and next frames to obtain the target information at the lane level for each target.

[0024] The beneficial effect is that real-time tracking of the target and corresponding information recognition can timely discover the target's driving status and provide an accurate data basis for subsequent congestion prediction.

[0025] Furthermore, in step three, when no abnormal driving behavior of large vehicles is detected, the reverse prediction model is used to automatically analyze abnormal traffic information, which includes short-term large fluctuations in traffic volume, a significant decrease in vehicle speed, and a significant increase in illegal lane changes, to predict traffic congestion.

[0026] The beneficial effect is that by making additional traffic congestion predictions for situations where abnormal driving behavior of large vehicles is not detected, potential traffic problems can still be detected even if abnormal driving of large vehicles is not clearly detected, thereby maintaining the smoothness and safety of the roads in a more comprehensive manner.

[0027] Furthermore, in step 4, when abnormal driving behavior of a large vehicle is not detected, the traffic information features are input into the reverse prediction model trained by deep learning for calculation to obtain a traffic congestion prediction result.

[0028] The beneficial effect is that the model is used to automatically predict traffic congestion, and the prediction results are more intelligent and reliable.

[0029] Furthermore, in step 4, a forward prediction model and a reverse prediction model are obtained by training an LSTM model using real-time updated historical data, and the traffic congestion prediction results include the time when congestion occurs, the duration of congestion, and the congestion level.

[0030] Furthermore, in step four, the LSTM model includes an input layer, an LSTM layer, and an output layer. The input layer includes the types of abnormal driving of large vehicles, the traffic volume of each lane, and the average speed of each lane. The LSTM layer is used to process time series data and capture temporal dependencies and changing trends. The output layer is used to output congestion predictions, including the time of congestion occurrence, congestion duration, and congestion level.

[0031] Furthermore, in step 4, the operation process of the LSTM model is as follows:

[0032] The input gate controls the degree of influence of the new information of the current time step on the cell state. The output of the input gate is expressed as:

[0033] ;

[0034] Among them, σ is the Sigmoid activation function, and is the weight matrix and bias term of the input gate, is the hidden state of the previous time step, is the input of the current time step;

[0035] Determine the cell state at the previous time step through the forget gate The information retained in the forget gate is expressed as:

[0036] ;

[0037] in, and are the weight matrix and bias term of the forget gate respectively;

[0038] Update the cell state according to the output of the input gate and the forget gate. First calculate the candidate cell state, which is expressed as:

[0039] ;

[0040] The output of the forget gate and the output of the input gate are used to update the cell state in a weighted manner, which is expressed as:

[0041] ;

[0042] Use the output gate to control the hidden state of the current time step , the output of the output gate is:

[0043] ;

[0044] in, and is the weight matrix and bias term of the output gate. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a system architecture diagram of a first embodiment of a method for predicting traffic congestion caused by abnormal driving of large vehicles according to the present invention;

[0046] Figure 2 This is a diagram of the LSTM network structure in Example 1 of the method for predicting traffic congestion caused by abnormal driving of large vehicles according to the present invention. DETAILED DESCRIPTION

[0047] The following is further explained in detail through specific implementation methods.

[0048] Example 1

[0049] Methods for predicting traffic congestion caused by abnormal driving of large vehicles, such as Figure 1 As shown, the following steps are included:

[0050] Step 1: Use cameras and radar equipment to collect road traffic information in real time. Road traffic information includes image information and point cloud information. Cameras and radar equipment can use common equipment for existing traffic data collection, which will not be described in detail here.

[0051] Step 2: Process the image and point cloud information to obtain target information for each object. Image processing involves using the YOLOv5 model in computer vision to identify the pixel coordinates and target categories of objects in the image. These categories include non-motorized vehicles, motorcycles, cars, buses, and trucks. Motorcycles also include bicycles. Point cloud processing involves preprocessing the point cloud information, performing registration, integrating point cloud data from multiple locations or time periods, extracting features to obtain edge and plane information from the point cloud, performing object segmentation and recognition, segmenting the point cloud into different objects, and classifying the target categories.

[0052] After feature-level fusion of the camera and radar equipment, the fusion adopts the following formula:

[0053] ;

[0054] in, is the fused feature. and Features from camera and radar respectively, is a weight parameter, with a value between 0 and 1.

[0055] The target is tracked and identified. The target information includes vehicle location, type, speed, heading angle and lane-level information. The target information also includes traffic flow and illegal driving information, among which illegal driving includes: illegal lane changing, illegal parking, and driving in the wrong direction.

[0056] The process of acquiring target information is as follows: tracking the target through the Deepsort model and identifying the motion trajectory of each target; the DeepSort (Deep Learning-based Object Tracking) model is a target tracking model based on deep learning, which is mainly used to accurately detect and track targets in video sequences. The DeepSort model combines feature extraction of deep learning with target tracking of Kalman filtering and Hungarian algorithm to achieve efficient and accurate detection and tracking of targets in video sequences. The key idea of ​​Kalman filtering lies in data fusion and iterative updating. By combining prior information and actual measurements, the system can better estimate the current state and gradually improve the accuracy of the estimation through iterative updates. Its core assumption is that the dynamic system is linear, which can be expressed as:

[0057] ;

[0058] ;

[0059] It learns the target appearance features through deep learning and combines Kalman filtering and Hungarian algorithm to estimate the target state, thus adapting to multi-target tracking scenarios.

[0060] The identified trajectory is mapped to the world coordinate system using the Homography matrix to show the target's true trajectory on the map. Homography is a 3x3 transformation matrix that maps points on an image to corresponding points in the world coordinate system. Its principle can be expressed as a 3x3 matrix H, expressed as:

[0061] ;

[0062] For a pair of corresponding points in the image, located at (x1, y1) in the image and (x2, y2) in the world coordinate system, the mapping relationship between the two is established through H, expressed as:

[0063] ;

[0064] For all corresponding points, a one-to-one mapping relationship between image pixel coordinates and world coordinates is established, so that the target detected by the camera can be displayed on the map.

[0065] Each target is directly analyzed in depth to obtain information including vehicle location, type, speed, etc. The Deepsort model is used to compare the positions of the targets in the previous and next frames to obtain target information at the lane level for each target. This data not only helps to monitor traffic flow in real time, but also provides comprehensive insights into vehicle behavior on the road. At the same time, the system also carefully monitors each element in the traffic flow and captures lane-level information such as traffic flow and illegal driving. The definition of illegal driving includes irregular behaviors such as illegal lane changes, illegal parking, and driving against the flow. The following is the specific information obtained:

[0066] Traffic flow, or the total number of vehicles passing through a given timeframe, is a key indicator of traffic flow. To provide more detailed traffic flow information, this invention breaks down traffic flow data and analyzes it by lane. This approach provides more accurate traffic flow information, enabling a more comprehensive understanding of intersection trends.

[0067] Traffic flow data collection method: Lane-level data collection is achieved through individual cameras. Specifically, a rectangular area is divided into two sections in each lane, with the stop line as one edge. When a target passes through this area, a counter is incremented by 1. The number of vehicles passing through per unit time is used as the traffic flow in that lane.

[0068] Vehicle speed: This refers to the speed of a vehicle, a crucial parameter for assessing traffic conditions and road flow. Speed ​​data is collected using a Deepsort tracking model, which tracks the target's movement trajectory and calculates its real-time speed. By monitoring vehicle speed changes in real time, potential congestion can be identified in advance.

[0069] Vehicle Classification: A more detailed classification system has been introduced, covering five major categories: non-motorized vehicles, bicycles, cars, buses, and trucks. This more granular classification allows for a more comprehensive understanding of the diversity of traffic participants, while also focusing on trucks. By gaining real-time information about trucks, potential traffic congestion caused by trucks can be detected in advance. This is crucial for predicting traffic congestion, as trucks are typically larger and have different driving characteristics, and their driving status has a significant impact on road capacity.

[0070] Illegal driving includes a range of irregular behaviors to more comprehensively monitor and evaluate the behavior of traffic participants. Specifically, illegal driving is categorized into three main areas: illegal lane changes, illegal parking, and wrong-way driving. Illegal lane changes encompass lane changes made in inappropriate locations or under inappropriate circumstances. This includes frequent lane changes in congested roads, intersections, or areas where lane changes are prohibited, leading to traffic disruption and danger. Illegal lane changes by large vehicles, in particular, can have a greater impact on traffic. We use real-time tracking to flag large vehicles when they change from one lane to another. Illegal parking refers to parking in no-parking zones, intersections, or other areas where parking is prohibited. This behavior can hinder normal traffic flow, cause congestion, and increase the risk of accidents. If real-time tracking indicates that a target's location information has not changed within three minutes, it is flagged as illegally parked and recorded to provide data support for subsequent assessments. Wrong-way driving refers to driving in a direction other than the designated direction, which can occur on one-way streets or other areas with restricted directional movement. Wrong-way driving not only violates traffic regulations but can also cause chaos at intersections and disrupt traffic flow. By tracking the target, if its trajectory is found to be moving in the opposite direction, it is marked as retrograde and the target is marked. The above data can provide data support for subsequent judgment.

[0071] Cameras capture visual information, while radar detects the vehicle's position and motion through reflected radio waves. By combining and fusing information from these two different sensors, the system achieves a more comprehensive and accurate understanding of traffic conditions. The fusion of image and lidar data enables more accurate target identification and tracking, capturing real-time traffic information with this fused data. This feature-level fusion not only improves the real-time nature of traffic information but also significantly enhances the system's perception and understanding of complex traffic scenarios, ensuring accurate target capture and laying a solid foundation for subsequent processing.

[0072] Step 3: Based on the road traffic information in step 1, positive prediction is performed to detect abnormal driving behaviors of large vehicles. Abnormal driving behaviors of large vehicles include large vehicles illegally entering the fast lane, resulting in overall speed restrictions in the fast lane, and large vehicles breaking down on the main road, causing congestion.

[0073] When abnormal driving behavior of large vehicles is not detected, the reverse prediction model is used to automatically analyze abnormal traffic information, including short-term large fluctuations in traffic volume, significant reductions in vehicle speed, and a sharp increase in illegal lane changes, to predict traffic congestion.

[0074] In step 4, when unusual large vehicle behavior is detected, traffic information features are fed into a deep learning-trained forward prediction model for calculation, resulting in a traffic congestion prediction. This forward prediction model is trained to accept traffic information features of detected unusual large vehicle behavior, and an LSTM model automatically calculates the likelihood of traffic congestion. The model inputs include the type of unusual large vehicle behavior, the traffic volume in each lane, and the average speed in each lane. The output is a congestion prediction, including the time of occurrence, duration, and level of congestion. Congestion levels range from mild (Level 1), moderate (Level 2), and severe (Level 3). Level 1 features describe a minor traffic congestion on the road section, with traffic flow slowing slightly, but overall speeds are maintained. Level 1 possible causes: Minor congestion may be caused by traffic incidents, heavy traffic at intersections, or brief vehicle decelerations. Level 2 features describe a significant impact on traffic flow on the road section, with vehicle speeds slowing, vehicle density increasing, and speeds decreasing significantly. Possible causes: Illegal vehicle driving, accidents, road construction and other reasons lead to obstruction of traffic flow, resulting in moderate congestion. Level 3 feature description: Traffic on the road section is completely blocked, vehicles are traveling very slowly, or even stagnant, forming a traffic jam. Possible causes: Serious accidents, road closures, large-scale vehicle influx and other reasons lead to the road section being unable to pass normally, resulting in severe congestion. The three-level congestion classification system can more accurately reflect the different levels of traffic conditions, and help traffic management departments take more targeted measures to alleviate congestion and improve road traffic efficiency. With the support of real-time monitoring and forward prediction models, the system can quickly and accurately identify different levels of congestion and provide a strong reference for traffic management decisions.

[0075] When undetected large vehicles exhibit unusual driving behavior, traffic information features are fed into a deep learning-trained reverse prediction model to generate traffic congestion predictions. By combining forward and reverse predictions, the system's sensitivity to unusual traffic behavior is enhanced, providing more comprehensive information support for traffic management and effectively predicting and addressing potential traffic issues.

[0076] like Figure 2 As shown, using real-time historical data, the forward prediction model and the reverse prediction model are trained through the LSTM model. The traffic congestion prediction results include the time of congestion occurrence, congestion duration, and congestion level. The LSTM model includes an input layer, an LSTM layer, and an output layer. The input layer includes the types of abnormal driving of large vehicles, the traffic volume of each lane, and the average speed of each lane. The LSTM layer is used to process time series data to capture temporal dependencies and changing trends. The output layer is used to output congestion predictions, including the time of congestion occurrence, congestion duration, and congestion level. The operation process of the LSTM model is as follows:

[0077] The input gate controls the degree of influence of the new information of the current time step on the cell state. The output of the input gate is expressed as:

[0078] ;

[0079] Among them, σ is the Sigmoid activation function, and is the weight matrix and bias term of the input gate, is the hidden state of the previous time step, is the input of the current time step;

[0080] Determine the cell state at the previous time step through the forget gate The information retained in the forget gate is expressed as:

[0081] ;

[0082] in, and are the weight matrix and bias term of the forget gate respectively;

[0083] Update the cell state according to the output of the input gate and the forget gate. First calculate the candidate cell state, which is expressed as:

[0084] ;

[0085] The output of the forget gate and the output of the input gate are used to update the cell state in a weighted manner, which is expressed as:

[0086] ;

[0087] Use the output gate to control the hidden state of the current time step , the output of the output gate is:

[0088] ;

[0089] in, and is the weight matrix and bias term of the output gate.

[0090] The forward and reverse prediction models utilize deep learning training using real-time historical data to adapt to the ever-changing dynamic traffic environment and driving behavior. This includes acquiring the latest traffic information and data on unusual large-scale vehicle behavior. By employing efficient optimization algorithms, both models ensure highly accurate traffic congestion predictions in a variety of scenarios. This dynamic training mechanism enables the forward and reverse prediction models to flexibly adapt to new driving behaviors and complex traffic scenarios. They can capture the evolution of the traffic system in real time and maintain sensitivity to traffic congestion issues. Through continuous real-time training and optimization, the two models not only provide reliable prediction capabilities but also lay a solid foundation for the reliability and stability of intelligent transportation systems. This approach has important practical application value for timely responding to diverse traffic management challenges and ensuring road traffic efficiency.

[0091] Compared with the existing technology, the solution of the first embodiment of this invention specifically extracts information related to target driving from the collected images and point cloud information, classifies it, distinguishes large vehicles, directly monitors abnormal driving of large vehicles, and makes positive predictions of traffic congestion based on abnormal driving behavior; when abnormal driving is not detected, it monitors abnormal traffic information and makes negative predictions of traffic congestion, monitoring different situations that cause traffic congestion in a classified manner. The amount of data processed is relatively small, and the processing time is relatively short, making it possible to quickly, promptly, and in real time monitor congestion within a limited time. Directly monitoring abnormal driving of large vehicles makes the monitoring results more accurate without using complex models, and improves the accuracy of traffic congestion predictions.

[0092] Example 2

[0093] The method for predicting traffic congestion caused by abnormal driving of large vehicles is different from the method of embodiment 1 in that in step 3, when positively predicting and detecting abnormal driving behavior of large vehicles, the target category of the large vehicle in the image information is identified. When the target category is a large vehicle, such as a truck or bus, the load information of the target is judged according to the target category. The identification of the target category can be performed through the vehicle outline. The load information includes underloaded, fully loaded, overloaded and empty. The load information is judged based on the size of the target's load object outline located within the original vehicle outline. Buses generally have regulations that they will not be overloaded and their general driving paths will not violate regulations. For example, if the load object outline is located inside the vehicle outline, the load information is underloaded. When the load object outline is located outside the vehicle outline, the load information is overloaded. When the load object outline is equal to the vehicle outline, the load information is fully loaded. When there is no load object outline, the load information is empty. According to the load information, the abnormal degree of the abnormal driving behavior of large vehicles is added. When the load information is empty, the abnormal degree of the abnormal driving behavior of the large vehicle is [0%-30%), when the load information is not fully loaded, the abnormal degree of the abnormal driving behavior of the large vehicle is [30%-50%), when the load information is fully loaded, the abnormal degree of the abnormal driving behavior of the large vehicle is [50%-70%), and when the load information is fully loaded, the abnormal degree of the abnormal driving behavior of the large vehicle is [70%-100%). The specific abnormal degree is determined according to the rated load of the large vehicle. The greater the rated load of the large vehicle, the greater the abnormal degree.

[0094] When abnormal driving behavior of a large vehicle is predicted and detected multiple times on a road section of preset length, an average value of the degree of abnormality is calculated. The preset length is set according to actual needs, for example, 1 kilometer. When the average value is greater than the threshold, it is judged as abnormal driving behavior of a large vehicle; when the average value is less than or equal to the threshold, the driving of the large vehicle is continuously tracked. If no traffic congestion occurs within the preset time, it is corrected to abnormal driving behavior of a non-large vehicle. The preset time can be set according to the actual judgment requirements, for example, the preset time is ten minutes.

[0095] Because large vehicles traveling on roads vary in size, and the specific load profiles of large vehicles of different sizes vary on actual roads, even if some large vehicles exhibit abnormal driving behavior, they may not cause congestion in some circumstances, leading to misjudgments of congestion. Therefore, after identifying abnormal driving behavior of large vehicles, this second embodiment determines the degree of abnormality of the abnormal driving behavior by evaluating load information. This allows for appropriate corrections to be made for different large vehicles of different target categories, improving the accuracy of subsequent congestion determinations.

[0096] The above is only an embodiment of the present invention, and the common knowledge such as the specific structure and characteristics of the scheme is not described in detail here. It should be pointed out that for those skilled in the art, several variations and improvements can be made without departing from the structure of the present invention, and these should also be regarded as the scope of protection of the present invention. These will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the description can be used to interpret the content of the claims.

Claims

1. A method for predicting traffic congestion caused by abnormal driving of large vehicles, characterized in that: The following steps are involved: Step 1: Use cameras and radar equipment to collect road traffic information in real time. Road traffic information includes image information and point cloud information; Step 2: Process the image and point cloud information to obtain target information for each target. This information is then fused at the feature level using the camera and radar equipment to track and identify the target. This target information includes vehicle location, type, speed, heading angle, and lane-level information. It also includes traffic flow and driving violations, including illegal lane changes, illegal parking, and driving against traffic. Step 3: Based on the road traffic information in step 1, positive prediction is performed to detect abnormal driving behaviors of large vehicles. Abnormal driving behaviors of large vehicles include large vehicles illegally entering the fast lane, resulting in overall speed restrictions in the fast lane, and large vehicles breaking down on the main road, causing congestion. When no abnormal driving behavior of large vehicles is detected, the reverse prediction model is used to automatically analyze abnormal traffic information, including short-term large fluctuations in traffic volume, significant reductions in vehicle speed, and significant increases in illegal lane changes, to predict traffic congestion. Step 4: When abnormal behavior of a large vehicle is detected, the traffic information features are input into the forward prediction model trained by deep learning for calculation to obtain the traffic congestion prediction result. When abnormal driving behavior of a large vehicle is not detected, the traffic information features are input into the reverse prediction model trained by deep learning for calculation to obtain the traffic congestion prediction result.

2. The method for predicting traffic congestion caused by abnormal driving of large vehicles according to claim 1, characterized in that: In the second step, the pixel coordinate information and target category of the target in the image information are identified by using the YOLOv5 model in computer vision, and the target categories include non-motor vehicles, motorcycles, cars, buses and trucks; Preprocess the point cloud information, then perform registration, integrate point cloud data from multiple locations or times, extract features to obtain edge and plane information in the point cloud, perform object segmentation and recognition, divide the point cloud into different objects, and classify the target categories.

3. The method for predicting traffic congestion caused by abnormal driving of large vehicles according to claim 2, characterized in that: In the second step, the target is tracked by the Deepsort model to identify the motion trajectory of each target; The identified trajectory is mapped to the world coordinate system through the Homography matrix to show the real motion trajectory of the target on the map; The Deepsort model is used to compare the positions of the targets in the previous and next frames to obtain the target information at the lane level for each target.

4. The method for predicting traffic congestion caused by abnormal driving of large vehicles according to claim 1, characterized in that: In the step 4, a forward prediction model and a reverse prediction model are obtained by training an LSTM model using real-time updated historical data. The traffic congestion prediction result includes the time when congestion occurs, the duration of congestion, and the congestion level.

5. The method for predicting traffic congestion caused by abnormal driving of large vehicles according to claim 4, characterized in that: In step 4, the LSTM model includes an input layer, an LSTM layer, and an output layer. The input layer includes the types of abnormal driving of large vehicles, the traffic volume of each lane, and the average speed of each lane. The LSTM layer is used to process time series data and capture temporal dependencies and changing trends. The output layer is used to output congestion predictions, including the time of congestion occurrence, congestion duration, and congestion level.

6. The method for predicting traffic congestion caused by abnormal driving of large vehicles according to claim 5, characterized in that: In step 4, the operation process of the LSTM model is as follows: The input gate controls the degree of influence of the new information of the current time step on the cell state. The output of the input gate is expressed as: ； Among them, σ is the Sigmoid activation function, and is the weight matrix and bias term of the input gate, is the hidden state of the previous time step, is the input of the current time step; Determine the cell state at the previous time step through the forget gate The information retained in the forget gate is expressed as: ； in, and are the weight matrix and bias term of the forget gate respectively; Update the cell state according to the output of the input gate and the forget gate. First calculate the candidate cell state, which is expressed as: ； The output of the forget gate and the output of the input gate are used to update the cell state in a weighted manner, which is expressed as: ； Use the output gate to control the hidden state of the current time step , the output of the output gate is: ； in, and is the weight matrix and bias term of the output gate.

Images