A lane-level driving assistance method and system based on traffic flow

Abstract

The present invention relates to a lane-level driving assistance method and system based on traffic flow. The method includes: determining a target section and obtaining traffic flow information of the target section; obtaining driving events detected during a vehicle's journey according to the traffic flow information, classifying the driving events based on an artificial neural network, and using fuzzy classification output to obtain the classified driving events; identifying extreme driving behaviors of a driver based on the traffic flow information and in combination with the classified driving events; obtaining a driving scenario and vehicle state changes based on the traffic flow information and the extreme driving behaviors; generating a target formation structure based on the driving scenario and the vehicle state changes; and performing vehicle target road allocation and motion planning based on the target formation structure to achieve the vehicle driving assistance function. Compared with the prior art, the present invention makes full use of environmental information and driving information to achieve assisted driving in a lane-level environment, and helps a driver find the best lane-level route on the premise of minimizing the journey time.

Description

Technical Field

[0001] The present invention relates to the technical field of autonomous driving for connected vehicles, and particularly to a lane-level driving assistance method and system based on traffic flow. Background Art

[0002] In recent years, advanced driving assistance systems based on wireless communication between vehicles and between vehicles and infrastructure have developed rapidly. The intelligent connected vehicle technology plays an important role in improving vehicle and road safety, efficiency, environmental sustainability, and driving comfort. However, existing advanced driving assistance systems can only rely on on-vehicle sensors to detect targets within a certain range and cannot coordinate conflicts and avoidance behaviors during vehicle interaction. In addition, although a large number of intelligent connected vehicle applications for driving assistance have been proposed and developed, few of them focus on lateral control assistance. Therefore, driving based on a vehicle-road collaborative system, which interconnects the real-time changing driving environment with the driving behavior of the vehicle and enables the vehicle to perform corresponding assisted driving behaviors for different lane-level environments, is a technical difficulty that needs to be urgently solved by those skilled in the art. Summary of the Invention

[0003] The purpose of the present invention is to overcome the defects of the above-mentioned existing technologies and provide a lane-level driving assistance method and system based on traffic flow. The present invention utilizes the wireless communication information input between vehicles and between vehicles and infrastructure obtained by the perception system, realizes the collaborative interaction of vehicle speed information guidance, lane allocation, and optimal lane selection functions, makes full use of environmental information and driving information, realizes assisted driving in the lane-level environment, and helps the driver find the optimal lane-level route on the premise of minimizing the travel time.

[0004] The purpose of the present invention can be achieved by the following technical solutions:

[0005] A lane-level driving assistance method based on traffic flow for assisting vehicle autonomous driving, comprising the following steps:

[0006] Determine the target section and obtain the traffic flow information of the target section, where the traffic flow information includes acceleration information and speed information;

[0007] According to the traffic flow information, obtain the driving events detected during the vehicle journey, classify the driving events based on an artificial neural network, and use fuzzy classification output to obtain the classified driving events;

[0008] Based on the traffic flow information and in combination with the classified driving events, identify the extreme driving behaviors of the driver;

[0009] Based on the traffic flow information and the extreme driving behaviors, obtain the driving scenario and vehicle state changes;

[0010] Generate a target formation structure based on the driving scenario and vehicle state changes;

[0011] Based on the target formation structure, perform vehicle target road assignment and motion planning to achieve vehicle driving assistance functions.

[0012] Further, the acceleration information acquisition includes the following steps:

[0013] Obtain vehicle inertial data by an inertial sensor, obtain vehicle acceleration data by an accelerometer, and obtain the real-time position information of the vehicle by GPS;

[0014] Pass the data obtained by the inertial sensor, accelerometer and GPS through a Kalman filter to obtain the acceleration information of the traffic flow.

[0015] Further, the speed information acquisition includes the following steps:

[0016] Obtain the real-time video of the vehicle surrounding environment by a camera;

[0017] Apply an adaptive background subtraction algorithm to the real-time video to separate the environment from the vehicle;

[0018] Remove noise and shadows from the data after separating the environment from the vehicle, and perform vehicle target tracking to track the vehicle target in each frame of the real-time video to calculate the speed information of the traffic flow;

[0019] The vehicle target tracking includes target segmentation, target marking, bounding box extraction and center extraction.

[0020] Further, the driving events include lateral driving events and longitudinal driving events;

[0021] Obtain the driving events detected during the vehicle journey, classify the driving events based on an artificial neural network, and use fuzzy classification output to obtain the classified driving events, including the following steps:

[0022] Set the classification thresholds for longitudinal acceleration and yaw rate to demarcate the beginning and end of longitudinal driving events and lateral driving events; the classification thresholds include acceleration, deceleration, turning, lane change and braking;

[0023] Based on the traffic flow information, obtain the continuous driving profiles of different vehicles in the target section;

[0024] According to the pre-generated and trained data, use unsupervised learning techniques to analyze and extract the driving events manipulated by acceleration and braking from all the identified driving events, calculate their percentage scores and classification thresholds, and classify the driving events based on an artificial neural network.

[0025] Furthermore, for the driving events combined with classification to identify the extreme driving behaviors of the driver, the following steps are included:

[0026] When the driver performs abnormal operations, the acceleration will change sharply; the abnormal operations include rapid acceleration, emergency braking, and rapid lane change;

[0027] In the events of rapid acceleration and emergency braking, the longitudinal acceleration of the vehicle will increase sharply and remain for a period of time. In the event of rapid lane change, the lateral acceleration of the vehicle will change sharply;

[0028] By detecting the extreme points of acceleration to segment the extreme driving behaviors of the driver, the occurrence of accidents in the target section is reduced.

[0029] Furthermore, use a random forest classifier to detect the driving events under each classification threshold, and discretely classify and output the driving events of acceleration, braking, and turning maneuvers;

[0030] For the extreme driving behaviors, use principal component analysis, stacked sparse autoencoders, and statistical features to extract features, and through the recognition model of a long short-term memory neural network, statistically analyze the traffic flow information, detect the extreme acceleration points and extreme deceleration points caused by the driver's abnormal operations, and identify the extreme driving behaviors.

[0031] Furthermore, for the driving scenarios and vehicle state changes to generate the target formation structure, the following steps are included:

[0032] The roadside unit of the target section or the roadside unit of the nearby section sends the wireless communication information between vehicles, between the vehicle and the infrastructure, and the target section condition information to the on-vehicle unit within their respective corresponding radio communication ranges. The target section condition information includes the driving scenarios and vehicle state changes for the on-vehicle unit to control the autonomous driving vehicle to execute the corresponding autonomous driving strategies;

[0033] The communicable vehicles detect the traffic state through the vehicle-built sensors, and the on-vehicle intelligent system sends the detected vehicle activity information and traffic state to the cloud database;

[0034] Based on the data information transmitted by the communicable vehicles to the cloud database, the application server predicts the traffic state through the prediction model;

[0035] The application server stores and broadcasts the traffic state prediction results to the vehicles equipped with the application program;

[0036] Construct a conflict-free relative path planning algorithm, define the conflict types of the formation vehicles, and based on the traffic state prediction results, adopt a waiting strategy or a path-changing strategy according to the conflict types of the vehicles, and plan a conflict-free path through iterative update of the allocation results to avoid potential collisions;

[0037] Build a lane-changing algorithm for real-time allocation to minimize the total number of lane changes for all vehicles. For the optimal problem of the best lane-level path for application vehicles, determine the lane segment space that a vehicle should occupy at a specific time, and set an allocation coordination algorithm to help coordinate lane changes.

[0038] Furthermore, the idea of the lane-changing algorithm for real-time allocation is as follows:

[0039] Generate target lanes according to the number of vehicles, and allocate vehicles to target lane positions one by one;

[0040] Define a cost function L(i,j) to represent the cost of allocating vehicle i to target j. The total cost of the allocation result of vehicle i is L(i); the cost is set to the Euclidean distance between the vehicle and the target lane.

[0041] Taking the number of lane changes as the cost, after generating target lanes on the road, calculate the number of lane changes for each vehicle to each target, and find the allocation result with the least total number of lane changes.

[0042] Furthermore, the vehicle target road allocation and motion planning are carried out to realize the vehicle driving assistance function, including the following steps:

[0043] The traffic coordinator processes the environmental information between vehicles and generates the trajectories connecting the vehicles;

[0044] Using the preview method, mark the trajectories of the vehicles with a series of road points, and the vehicles track the road points to keep driving in the center of the lane or perform lane-changing maneuvers;

[0045] The vehicle performs trajectory planning with spatio-temporal constraints and optimal trajectory tracking, and the lane-changing control strategy of the vehicle integrates the control of formation keeping and desired speed keeping.

[0046] A lane-level driving assistance system based on traffic flow is used to implement the above-mentioned lane-level driving assistance method based on traffic flow, including an image recognition system, a gateway, a global dynamic map, a cloud database, a traffic coordinator, and an application server;

[0047] The image recognition system includes a camera and a processor. The camera is used to obtain real-time videos of the vehicle's surrounding environment, and software for image processing of the real-time videos is deployed in the processor;

[0048] The image recognition system communicates with the gateway and sends the environmental information between vehicles to the global dynamic map;

[0049] The global dynamic map is used to store traffic flow information and environmental information between vehicles;

[0050] The cloud database is used to store vehicle activity information and traffic status information detected by the vehicle intelligent system; the cloud database is connected to the application server;

[0051] The application server is used to predict the traffic status and store and broadcast the traffic status prediction results to vehicles equipped with the application program;

[0052] The traffic coordinator is used to process the environmental information between vehicles and generate the trajectories of connected vehicles;

[0053] The traffic coordinator includes a longitudinal controller and a lateral controller. The longitudinal controller is used to form and maintain the generated formation while maintaining the required speed; the lateral controller is used to perform lane keeping and lane changing operations.

[0054] Compared with the prior art, the present invention has the following beneficial effects:

[0055] By obtaining the input of wireless communication information between vehicles and between vehicles and infrastructure, and judging the conditions of the target section based on traffic flow information, the present invention realizes the collaborative interaction of speed information guidance and lane allocation functions, makes full use of environmental information and driving information, realizes lane-level assisted driving, helps the driver find the best lane-level route on the premise of minimizing the travel time, and at the same time provides conditions for the control of formation keeping and desired speed keeping. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 is a flowchart of the present invention;

[0057] Figure 2 is a flowchart of traffic flow information acquisition of the present invention;

[0058] Figure 3 is a flowchart of extreme driving behavior judgment in the present invention;

[0059] Figure 4 is a flowchart of lane-level driving assistance in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0060] The present invention will be described in detail below with reference to the drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and gives the detailed implementation manner and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0061] This embodiment proposes a lane-level driving assistance system based on traffic flow, including an image recognition system, a gateway, a global dynamic map, a cloud database, a traffic coordinator and an application server;

[0062] The image recognition system includes a camera and a processor. The camera is used to obtain real-time videos of the vehicle's surrounding environment, and software for image processing of the real-time videos is deployed in the processor.

[0063] The image recognition system communicates with the gateway and sends the environmental information between vehicles to the global dynamic map.

[0064] The global dynamic map is used to store traffic flow information and environmental information between vehicles.

[0065] The cloud database is used to store vehicle activity information and traffic status information detected by the vehicle-mounted intelligent system; the cloud database is connected to the application server.

[0066] The application server is used to predict the traffic status and store and broadcast the traffic status prediction results to vehicles equipped with applications.

[0067] The traffic coordinator is used to process the environmental information between vehicles and generate the trajectories of connected vehicles.

[0068] The traffic coordinator includes a longitudinal controller and a lateral controller. The longitudinal controller is used to form and maintain the generated formation while maintaining the required speed; the lateral controller is used to perform lane keeping and lane change operations.

[0069] Based on this system, a lane-level driving assistance method based on traffic flow can be realized for assisting vehicle autonomous driving, including the following steps:

[0070] S1. Determine the target section, obtain the traffic flow information of the target section, and upload it to the global dynamic map. The traffic flow information includes acceleration information and speed information.

[0071] S2. According to the traffic flow information, obtain the driving events detected during the vehicle's journey, classify the driving events based on an artificial neural network, and use fuzzy classification output to obtain the classified driving events.

[0072] S3. Based on the traffic flow information, combined with the classified driving events, identify the driver's extreme driving behaviors.

[0073] S4. Based on the traffic flow information and extreme driving behaviors, obtain the driving scenarios and vehicle state changes.

[0074] S5. Based on the driving scenarios and vehicle state changes, generate the target formation structure.

[0075] S6. Based on the target formation structure, perform vehicle target road allocation and motion planning to achieve the vehicle driving assistance function.

[0076] This embodiment performs vehicle detection based on traffic flow information, calculates the vehicle tracking speed, and determines the condition of the target road section. Since the vibrations inside the vehicle during driving will affect the data recorded by in-vehicle sensors with noise, in order to obtain more accurate speed and acceleration information to reflect the condition of the target road section, in this embodiment, the acceleration information of the vehicle is integrated by inertial sensor data, accelerometer data, GPS, and a Kalman filter.

[0077] As Figure 2 shown, specifically, the acquisition of acceleration information includes the following steps:

[0078] (1) Obtain the vehicle inertial data by the inertial sensor, obtain the vehicle acceleration data by the accelerometer, and obtain the real-time position information of the vehicle by GPS;

[0079] (2) Pass the data obtained by the inertial sensor, accelerometer, and GPS through the Kalman filter to obtain the acceleration information of the traffic flow.

[0080] The acquisition of speed information includes the following steps:

[0081] (1) Obtain the real-time video of the vehicle surrounding environment by the camera in the image recognition system;

[0082] (2) Apply the adaptive background subtraction algorithm to the real-time video to separate the environment from the vehicle;

[0083] (3) Remove noise and shadows from the data after separating the environment from the vehicle, and perform vehicle target tracking. Vehicle target tracking includes target segmentation, target marking, bounding box extraction, center extraction, etc. Track the vehicle target in each frame of the real-time video to calculate the speed information of the traffic flow;

[0084] The image recognition system communicates with the V2X gateway, forwards each message to the global dynamic map, and stores the environmental information about connected and unconnected vehicles in the database.

[0085] This embodiment analyzes the driving behavior of the driver and extends the recognition of traffic congestion conditions on the highway at the driver level. The discrete driving behavior selection of the vehicle has an important impact on helping to improve road capacity and reduce collision risks. In view of this, this embodiment takes the traffic flow information obtained from the above global dynamic map as the input, detects the models of longitudinal and lateral driving maneuvers, classifies several driving styles based on the driving events detected during the journey by an artificial neural network, which can be used for real-time driver monitoring and assistance systems, and extends the model to recognize traffic congestion conditions on the highway to better achieve the road traffic efficiency.

[0086] Among them, the driving events include lateral driving events and longitudinal driving events;

[0087] Obtain the driving events detected during the vehicle journey, classify the driving events based on an artificial neural network, and use fuzzy classification output to obtain the classified driving events, including the following steps:

[0088] Set the classification thresholds for longitudinal acceleration and yaw rate to demarcate the beginning and end of longitudinal driving events and lateral driving events; the classification thresholds include acceleration, deceleration, turning, lane change, and braking;

[0089] The characteristic parameters of driving events are characterized by a set of statistical values such as longitudinal acceleration, lateral acceleration, yaw rate, and event length, as Figure 3 shown;

[0090] Since the neural network determines its structure based on the experience of a given number of input and output neurons, the classification accuracy is very high, the calculation time of the artificial neural network is relatively short, and it allows fuzzy classification output, which can provide more information for driving style classification and distinguish more events.

[0091] In view of this, the present invention obtains the continuous driving profiles of different vehicles in a given section through the traffic flow information of the global dynamic map, analyzes and extracts the driving events manipulated by acceleration and braking from all the identified driving events using unsupervised learning techniques according to the pre-generated and trained data, calculates their score percentages and classification thresholds, and classifies the driving events based on an artificial neural network, aiming to identify different patterns of longitudinal acceleration and braking maneuvers exhibited by the driver.

[0092] Combined with the classified driving events, identify the extreme driving behaviors of the driver, including the following steps:

[0093] During driving, the acceleration and speed of the vehicle have only small fluctuations in a certain small interval most of the time, approximately in a steady state. When the driver performs abnormal operations, the acceleration will change sharply; the abnormal operations include rapid acceleration, emergency braking, and rapid lane change;

[0094] In rapid acceleration and emergency braking events, the longitudinal acceleration of the vehicle will increase sharply and remain for a period of time. In rapid lane change events, the lateral acceleration of the vehicle will change sharply; therefore, by detecting the extreme acceleration points, the extreme driving behaviors of the driver are segmented to reduce the occurrence of accidents in the target section.

[0095] Use a random forest classifier to detect the driving events under each classification threshold and discretely classify and output the driving events of acceleration, braking, and turning maneuvers;

[0096] For extreme driving behaviors, features are extracted using principal component analysis, stacked sparse autoencoders, and statistical features, and through an identification model of long short-term memory neural networks, statistical analysis is performed on traffic flow information to detect extreme acceleration points and extreme deceleration points caused by abnormal operations performed by drivers, and to identify extreme driving behaviors, including rapid acceleration events, emergency braking events, and rapid lane change events, in order to reduce the occurrence of accidents on the target road section.

[0097] As Figure 4 shown, it is a lane-level driving assistance flowchart. Based on the changes in the driving scenario in the global dynamic map and the data collected by vehicles with road communication capabilities, the application server or traffic management center predicts the traffic state and the vehicle switches the formation structure.

[0098] The roadside unit of the target road section or the roadside unit RSU of the nearby road section sends the wireless communication information between vehicles, between vehicles and infrastructure, and the target road section condition information obtained above to the on-board unit OBU within their respective corresponding radio communication ranges, for the on-board unit OBU to control the autonomous vehicle to execute the corresponding autonomous driving strategy.

[0099] Send messages in an end-to-end manner through the cellular network, and vehicles equipped with applications always have communication capabilities.

[0100] Communicable vehicles can send their activity information to a centralized library through the in-vehicle intelligent system, and built-in sensors are used for traffic state prediction.

[0101] Based on the data collected by vehicles with road communication capabilities, the application server or traffic management center can estimate and even predict the traffic state through a prediction model.

[0102] The application server stores and broadcasts the prediction results to vehicles equipped with applications, so that the recommended applications can support the decision-making process for lane changes.

[0103] The traffic state changes over time. When the number of lanes or the number of vehicles changes, a high level of flexibility in formation adjustment is required, so a dynamic model is needed to predict the traffic state.

[0104] The influencing factors of the decision-making process are divided into human factors, traffic environment, and road infrastructure elements.

[0105] There is a complex interaction between human drivers and the traffic environment. Most existing multi-vehicle coordination control methods focus on only considering the longitudinal behavior of vehicles in a single-lane traffic environment. The present invention takes into account that drivers will consider the lane environment and the potential influence of multiple surrounding vehicles when making driving decisions, rather than only considering the influence between a single leading vehicle and the host vehicle, and expands the driver behavior model.

[0106] To improve the efficiency and safety of lane changes, vehicles directly in front of the host vehicle, as well as those in the front - side and side positions of the left and right lanes, i.e., the staggered structure where vehicles in adjacent lanes drive longitudinally in a staggered manner, are considered in the modeling framework. Multi - vehicle multi - lane formation control is adopted, and the formation structure is smoothly and effectively switched according to the changing driving scenarios to make full use of the lane passing capacity. The potential influencing factors of the host vehicle are subdivided into the spatial distance and relative speed with multiple surrounding vehicles, and these variables are included as explanatory variables in the modeling framework, focusing on the global coordination of vehicles.

[0107] The characteristics of driving decisions are a combination of discrete and continuous components. In the present invention, joint modeling of discrete and continuous decisions is carried out, with the discrete and continuous components considered separately, that is, the decision to accelerate or decelerate or maintain the same speed and the degree of these decisions are separated, and discrete decisions and continuous decisions are processed separately.

[0108] The discrete component involves decisions to accelerate, decelerate, or maintain a constant speed, while the continuous component involves decisions about the degree of acceleration or deceleration. Truncated distributions of the acceleration and deceleration ranges are used to identify the physical and safety limits within which a driver can accelerate or decelerate, and interference from extreme acceleration values can be avoided.

[0109] The main purpose of speed decision - making is to establish a TCP connection between the roadside unit (RSU) of the target road section or the RSU of the nearby road section to send a speed guidance request.

[0110] Simulating a real - world platoon cooperative driving communication scenario, the on - board unit (OBU) broadcasts its own driving state to other vehicles periodically as background communication. The RSU calculates the recommended speed at which the OBU can pass without stopping based on the received request message containing vehicle speed and position information and the current signal state, and the guidance speed does not exceed a certain range of the OBU's current vehicle speed, with a corresponding change amount specified.

[0111] In view of this, the present invention proposes a motion planning framework aimed at the smoothness and safety of platoon vehicle switching; at the same time, a conflict - free relative path planning algorithm is proposed, which defines the conflict types of platoon vehicles, and adopts a waiting strategy or a path - changing strategy according to the conflict types, and plans a conflict - free path through iterative update of the allocation result, avoiding potential collisions. When the number of lanes changes, the geometric staggered structure of the platoon will change.

[0112] This embodiment considers the formation adjustment function when new vehicles join the platoon and in - platoon vehicles leave the platoon. To allocate vehicles to the target positions in the generated formation, the present invention proposes a real - time allocation lane - changing algorithm to minimize the total number of lane - changes of all vehicles, determines the lane - segment space that a vehicle should occupy at a specific time for the optimal problem of the best lane - level path of the application vehicle, and sets an allocation coordination algorithm to assist in coordinating lane - changes.

[0113] Its basic idea is to generate targets according to the number of vehicles and then assign the vehicles to the target lane positions one by one. By defining a cost function L(i, j) to represent the cost of assigning vehicle i to target j, the total cost of the assignment result of vehicle i is L(i), aiming to select an appropriate cost function L(i, j). The cost is set to be the Euclidean distance between the vehicle and the target, which can minimize the total relative displacement and ensure that any two non-collinear trajectories do not overlap, thereby reducing the potential collision risk. Taking the number of lane changes as the cost, after generating targets on the road, the number of lane changes for each vehicle to each target can be calculated, and the assignment result with the least total number of changed lanes can be found using this algorithm.

[0114] The traffic coordinator will process the environmental data of connected and unconnected vehicles to generate the trajectories of connected vehicles. Using the preview method, the trajectories of vehicles are marked by a series of road points, and the vehicles need to track the road points to stay in the center of the lane or perform lane-changing maneuvers. The vehicles perform trajectory planning with spatio-temporal constraints and optimal trajectory tracking. The lane-changing control strategy of the vehicles integrates the control of formation keeping and desired speed keeping. The longitudinal controller is used to form and maintain the generated formation while maintaining the required speed. The lateral controller is used to perform lane keeping and lane-changing maneuvers.

[0115] This embodiment is based on the interactive driving behaviors of highway driving vehicles in the real world, and uses the simulation software Carla to construct a simulation scenario to verify the lane-level driving assistance method based on traffic flow proposed in this paper.

[0116] The preferred specific embodiments of the present invention have been described in detail above. It should be understood that those of ordinary skill in the art can make many modifications and variations according to the concept of the present invention without creative labor. Therefore, all technical solutions that can be obtained by those skilled in the art in the technical field of the present invention based on the concept of the present invention through logical analysis, reasoning or limited experiments on the basis of the prior art should be within the protection scope determined by the claims.

Claims

1. A lane-level driving assistance method based on traffic flow, characterized in that For assisting vehicle autonomous driving, including the following steps: Determine the target road section and obtain the traffic flow information of the target road section, where the traffic flow information includes acceleration information and speed information; According to the traffic flow information, obtain the driving events detected during the vehicle's journey, classify the driving events based on an artificial neural network, and use fuzzy classification output to obtain the classified driving events; Based on the traffic flow information and combined with the classified driving events, identify the driver's extreme driving behaviors; Based on the traffic flow information and extreme driving behaviors, obtain the driving scenario and vehicle state changes; Based on the driving scenario and vehicle state changes, generate the target formation structure; Based on the target formation structure, perform vehicle target road assignment and motion planning to achieve the vehicle driving assistance function; The driving events include lateral driving events and longitudinal driving events; Obtain the driving events detected during the vehicle's journey, classify the driving events based on an artificial neural network, and use fuzzy classification output to obtain the classified driving events, including the following steps: Set the classification thresholds for longitudinal acceleration and yaw rate to divide the beginning and end of longitudinal driving events and lateral driving events; the classification thresholds include acceleration, deceleration, turning, lane change, and braking; Based on the traffic flow information, obtain the continuous driving profiles of different vehicles in the target road section; According to the data generated and trained in advance, use unsupervised learning techniques to analyze and extract the driving events manipulated by acceleration and braking from all the identified driving events, calculate their score percentages and classification thresholds, and classify the driving events based on an artificial neural network; The driving scenario and vehicle state changes to generate the target formation structure, including the following steps: The roadside unit of the target road section or the roadside unit of the nearby road section sends the wireless communication information between vehicles, between vehicles and infrastructure, and the target road section condition information to the on-vehicle unit within their respective corresponding radio communication ranges, where the target road section condition information includes the driving scenario and vehicle state changes; The communicable vehicles detect the traffic state through the vehicle-built sensors, and the on-vehicle intelligent system sends the detected vehicle activity information and traffic state to the cloud database; Based on the data information transmitted by the communicable vehicles to the cloud database, the application server predicts the traffic state through a prediction model; The application server stores and broadcasts the traffic state prediction results to the vehicles equipped with the application program; Construct a conflict-free relative path planning algorithm, define the conflict types of the formation vehicles, and based on the traffic state prediction results, adopt a waiting strategy or a path-changing strategy according to the conflict types of the vehicles, and plan a conflict-free path through iterative update of the allocation results to avoid potential collisions; Construct a real-time allocation lane-changing algorithm to minimize the total number of lane changes of all vehicles. For the optimal problem of the best lane-level path of the application vehicle, determine the lane segment space that the vehicle should occupy at a specific time, and set an allocation coordination algorithm to help coordinate lane changes; The idea of the real-time allocation lane-changing algorithm is: Generate the target lanes according to the number of vehicles and allocate the vehicles to the target lane positions one by one; Define the cost function \(L(i, j)\), which represents the cost of vehicle \(i\) assigned to target \(j\), and the total cost of the assignment result of vehicle \(i\) is \(L(i)\); the cost is set to the Euclidean distance between the vehicle and the target lane. Taking the number of lane changes as the cost, after generating the target lanes on the road, calculate the number of lane changes for each vehicle to each target, and find the assignment result with the least total number of lane changes.

2. The lane-level driving assistance method based on traffic flow according to claim 1, wherein The acquisition of the acceleration information includes the following steps: Obtain the vehicle inertial data by the inertial sensor, obtain the vehicle acceleration data by the accelerometer, and obtain the real-time position information of the vehicle by the GPS; Pass the data obtained by the inertial sensor, accelerometer and GPS through the Kalman filter to obtain the acceleration information of the traffic flow.

3. A lane-level driving assistance method based on traffic flow according to claim 1, characterized in that, The acquisition of the speed information includes the following steps: Obtain the real-time video of the vehicle surrounding environment by the camera; Apply the adaptive background subtraction algorithm to the real-time video to separate the environment from the vehicle; Remove noise and shadows from the data after separating the environment from the vehicle, and perform vehicle target tracking, tracking the vehicle target in each frame of the real-time video to calculate the speed information of the traffic flow; The vehicle target tracking includes target segmentation, target marking, bounding box extraction and center extraction.

4. A lane-level driving assistance method based on traffic flow according to claim 1, characterized in that, The combination of classified driving events to identify the extreme driving behaviors of the driver includes the following steps: When the driver performs an abnormal operation, the acceleration will change sharply; the abnormal operations include rapid acceleration, emergency braking and rapid lane change; In the events of rapid acceleration and emergency braking, the longitudinal acceleration of the vehicle will increase sharply and remain for a period of time. In the event of rapid lane change, the lateral acceleration of the vehicle will change sharply; Segment the extreme driving behaviors of the driver by detecting the extreme acceleration points to reduce the occurrence of accidents on the target section.

5. The lane-level driving assistance method based on traffic flow according to claim 4, wherein, Use the random forest classifier to detect the driving events under each classification threshold, and discretely classify and output the driving events of acceleration, braking and turning maneuvers; For the extreme driving behaviors, use principal component analysis, stacked sparse autoencoders and statistical features to extract features, and through the recognition model of the long short-term memory neural network, statistically analyze the traffic flow information, detect the extreme acceleration points and extreme deceleration points caused by the driver's abnormal operations, and identify the extreme driving behaviors.

6. The lane-level driving assistance method based on traffic flow according to claim 1, wherein, The vehicle target road assignment and motion planning are performed to implement the vehicle driving assistance function, including the following steps: The traffic coordinator processes the environmental information between vehicles and generates the trajectories connecting the vehicles; Using the preview method, mark the trajectories of the vehicle by a series of road points, and the vehicle tracks the road points to stay in the center of the lane or perform lane change maneuvers; The vehicle performs trajectory planning with spatio-temporal constraints and optimal trajectory tracking, and the lane change control strategy of the vehicle integrates the control of formation keeping and desired speed keeping.

7. A lane-level driving assistance system based on traffic flow, which is used to implement a lane-level driving assistance method based on traffic flow as described in any one of claims 1-6, characterized in that, Including an image recognition system, a gateway, a global dynamic map, a cloud database, a traffic coordinator and an application server; The image recognition system includes a camera and a processor. The camera is used to obtain the real-time video of the vehicle surrounding environment, and the software for image processing of the real-time video is deployed in the processor; The described image recognition system communicates with the gateway and sends the environmental information between vehicles to the global dynamic map; The global dynamic map is used to store traffic flow information and environmental information between vehicles; The cloud database is used to store vehicle activity information and traffic status information detected by the in-vehicle intelligent system; The cloud database is connected to the application server; The application server is used to predict the traffic status and store and broadcast the traffic status prediction results to vehicles equipped with the application; The traffic coordinator is used to process the environmental information between vehicles and generate the trajectories of connected vehicles; The traffic coordinator includes a longitudinal controller and a lateral controller. The longitudinal controller is used to form and maintain the generated formation while maintaining the required speed. The lateral controller is used to perform lane keeping and lane changing operations.

Images