Method, safety helmet and system for ghost probe traffic accident early warning

Abstract

The invention belongs to the field of traffic accident early warning, and provides a method, a safety helmet and a system for ghost probe traffic accident early warning, and the method comprises the steps: obtaining an interaction sequence based on the data obtained by an electric vehicle rider through employing an SAN-GAN algorithm, and obtaining a ghost probe traffic accident early warning result through employing a pre-trained neural network model; generating hidden feature information of the electric vehicle rider according to the interaction sequence, and predicting a future trajectory of the electric vehicle rider according to the hidden feature information; based on the predicted future trajectory, utilizing an uncertainty position prediction method, introducing a scale factor, and calculating and predicting a collision area between the electric vehicle rider and the dangerous vehicle; based on the collision area, longitudinal and transverse collision danger state judgment is conducted on the electric vehicle rider according to the collision danger state evaluation index, a danger state judgment result is obtained, and early warning information is formed. According to the invention, early warning of ghost probe accidents is realized.

Description

Technical Field

[0001] The invention belongs to the field of traffic accident early warning, and in particular relates to a method, a safety helmet and a system for early warning of ghost-heading traffic accidents. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Among the numerous traffic accidents that cause injuries and deaths to electric bike riders, ghost-peek accidents are a common type of accident on urban roads. Ghost-peek accidents refer to an accident in which a motor vehicle (dangerous vehicle) and a non-motor vehicle (electric bike rider) appear from outside the sight triangle of each other at close range and in a short time due to the influence of a stopped obstructing vehicle, leaving neither party with time and space to react or evade, and eventually a collision occurs.

[0004] Most existing studies use motor vehicles as the main body of warning, and use fixed roadside facilities as the carrier of warning equipment. There is a lack of research on active collision prevention for electric bike riders. At the same time, there are certain limitations in the angle and accuracy of accident warning. Therefore, it is particularly important to provide an invention that can effectively protect the safety of electric bike riders, reduce the incidence of ghost-peek accidents, and has practical feasibility in solving urban traffic safety problems. Summary of the invention

[0005] In order to solve the above problems, the present invention proposes a method, a safety helmet and a system for warning of ghost-headed traffic accidents. The present invention combines the communication technology of information interconnection between vehicles and electric vehicle riders (referred to as vehicle-to-vehicle V2P communication) and the global positioning system (GPS) and other technologies, extracts information such as the time and position coordinates of dangerous collisions as accident characteristic variables under the normal operation of electric vehicles, and uses the long short-term memory network (LSTM) and collision danger state evaluation index based on the pedestrian trajectory prediction algorithm (SAN-GAN algorithm) to achieve early warning of ghost-headed accidents.

[0006] According to some embodiments, the present invention adopts the following technical solutions: A method for early warning of a ghost-peek traffic accident comprises the following steps: Obtain the driving environment, vehicle distance, traffic flow, coordinate position and speed direction of the electric vehicle rider; Obtain the position, speed and direction of each motor vehicle within a certain range of the electric vehicle rider's current driving position; Using the SAN-GAN algorithm, based on the data obtained from the electric bike rider, the interaction sequence is obtained. Using the pre-trained neural network model, the hidden feature information of the electric bike rider is generated according to the interaction sequence, and the future trajectory of the electric bike rider is predicted based on the hidden feature information. Based on the predicted future trajectory, the uncertainty position prediction method is used to introduce a proportional factor to calculate and predict the collision area between the electric bike rider and the dangerous vehicle; Based on the collision area, the longitudinal and lateral collision danger state of the electric vehicle rider is judged according to the collision danger state evaluation index, the danger state judgment result is obtained, and early warning information is formed.

[0007] As an optional implementation, the training process of the neural network model includes: collecting images in the target area, obtaining the trajectory of each electric bike rider and the deflection angle of the rider's head at regular intervals, and calibrating them to define the two-dimensional coordinates of the i-th electric bike rider ( , ) and its head deflection angle , multiple sets of effective electric vehicle rider trajectories and head deflection angles are obtained, at least a part of them are taken as training data, and after being processed by the SAN-GAN algorithm, the interaction sequence of each electric vehicle rider is obtained to train the neural network model.

[0008] As an optional implementation, the process of generating hidden feature information of an electric vehicle rider according to an interaction sequence includes: using a trajectory generator to generate hidden feature information of an electric vehicle rider, which includes a trajectory encoder, an electric vehicle rider field pooling module and a trajectory decoder, wherein the trajectory encoder is used to encode to obtain a feature vector of a target electric vehicle rider, and obtain a feature vector of a historical trajectory of the electric vehicle rider; The electric vehicle rider field of view pooling module is used to establish the electric vehicle rider field of view based on the interaction relationship between the electric vehicle riders, establish the electric vehicle rider field of view function, and obtain the electric vehicle rider field of view hidden feature vector; The relative distance between the target electric bike rider and other electric bike riders is passed through a fully connected layer to obtain a feature vector, which is combined with the hidden feature vector of the electric bike rider’s field of view and used as the input of the trajectory decoder; The trajectory decoder is used for decoding to obtain the trajectory prediction point of the electric vehicle rider.

[0009] As a further implementation method, the process of establishing the electric vehicle rider's field of view based on the interaction relationship between the electric vehicle riders includes: combining the actual change state of the electric vehicle rider's field of view during the interaction process to establish the electric vehicle rider's field of view : ; ; ; in, is the visual field of the i-th electric bike rider at time t; for The polar diameter; l is the length of the angle bisector that bisects the area of ​​the field of view; For this moment the polar angle of the electric bike rider; is the head deflection angle of the electric bike rider at this moment, that is l Angle with the horizontal axis; The actual field of view of the electric bike rider. is the relative distance between electric bike riders i and j.

[0010] As a further implementation method, the process of establishing the electric vehicle rider's field of view function and obtaining the electric vehicle rider's field of view hidden feature vector includes: considering that multiple other electric vehicle riders are simultaneously in the field of view of the predicted electric vehicle rider, taking the predicted electric vehicle rider as the influence source, the closer the electric vehicle rider in the field of view is to the influence source, the greater the repulsive force it generates, and the farther the distance is, the smaller the repulsive force generated, and accordingly establishing the electric vehicle rider's field of view function , and obtain the impact of the position information on the electric bike rider : ; ; In the formula, is the repulsion amplitude parameter between electric bike riders; r is the minimum collision-free distance between electric bike riders; represents the visual range of electric bike rider i at time t; is the field of view function of the electric bike rider; for The judgment parameters in .

[0011] As an optional implementation, the process of calculating and predicting the collision area between an electric bike rider and a dangerous vehicle includes: using the neural network model to obtain the mean and covariance of the future trajectory of the electric bike rider, assuming that the probability of the electric bike rider being located at a certain position depends on the collision area Ri, Ri is determined based on the trajectory mean and covariance matrix at time t, and a unit circle is mapped using the square root of the covariance matrix to determine an ellipse whose major semi-axis depends on the eigenvalues ​​of the matrix and whose direction is related to the corresponding eigenvector; using the inverse Mahalanobis transform, different proportional factors are introduced to calculate the confidence probability of any ellipse, and the area of ​​the collision area is represented by a confidence ellipse, and the closer to the center position of the ellipse, the higher the collision probability.

[0012] As an optional implementation, the collision danger status evaluation index includes the time it takes for the dangerous vehicle to reach the dangerous collision area in the longitudinal direction, the time it takes for the electric vehicle rider to reach the dangerous collision area in the longitudinal direction, the time it takes for the electric vehicle rider to enter the dangerous collision area in the lateral direction relative to the dangerous vehicle, and the time it takes for the electric vehicle rider to leave the dangerous collision area in the lateral direction relative to the dangerous vehicle.

[0013] As an optional implementation method, before making a judgment on the longitudinal and lateral collision danger state, the distance and angle between the electric vehicle rider and the blocking vehicle are obtained, the blocking vehicle is used as the coordinate origin of the coordinate system, a coordinate system is constructed, and the electric vehicle rider's position coordinate system is converted to determine the coordinates of the electric vehicle rider in the global coordinate system. Based on the coordinates of the electric vehicle rider in the global coordinate system and the coordinates of the dangerous vehicle in the global coordinate system, the relative position relationship between the dangerous vehicle and the electric vehicle rider is determined, and then the longitudinal speed and lateral speed of the electric vehicle rider relative to the vehicle coordinate system of the dangerous vehicle are calculated.

[0014] As an optional implementation, the process of determining the longitudinal collision danger state includes: establishing a longitudinal collision danger state prediction method based on the acquired motion state information and coordinate position information of the electric vehicle rider, as well as the motion state information and coordinate position information of the dangerous vehicle, combined with the prediction of the trajectory of the electric vehicle rider, and taking into account the time it takes for the vehicle to maintain the current speed to reach the collision point. Longitudinal dangerous state judgment model: ; In the formula, are the relative speed and relative acceleration of the dangerous vehicle and the electric bike rider respectively; D is the longitudinal distance between the dangerous vehicle and the electric bike rider at the collision position; L is the length of the dangerous vehicle; is the shortest time for an electric bike rider to reach the dangerous collision area, predicted by the neural network model; Compare and The size of > It is a safe state when the vehicle is in the forward collision state, otherwise, it is a dangerous state of longitudinal collision.

[0015] The process of judging the lateral collision danger state includes: the time when the electric bike rider enters and leaves the dangerous collision area , for: ; ; In the formula, is the width of the dangerous vehicle; is the width of the electric bike rider; is the minimum safe distance that the dangerous vehicle and the electric bike rider need to maintain; S is the lateral distance at the collision position between the dangerous vehicle and the electric bike rider; is the lateral speed of the electric bike rider relative to the coordinate system of the dangerous vehicle; According to the calculation , compare the size relationship between the three, when When , it is safe; when When , it is safe; when It is a lateral dangerous state.

[0016] As an optional implementation, the process of obtaining the dangerous state judgment result and forming the warning information includes: if both the longitudinal and lateral dangerous states are dangerous states, it is judged that there is a risk of collision and a warning is issued.

[0017] A safety helmet for early warning of ghost-heading traffic accidents, comprising a safety helmet body, on which a camera, a microwave sensor, a positioning module, a wireless communication module, a voice module and a control module are arranged, wherein: The camera is arranged in front of the safety helmet body and is used to obtain dynamic image information of the road surface; The microwave sensor is arranged in front of the safety helmet body and is used to obtain vehicle flow, instantaneous vehicle speed and vehicle distance information; The positioning module is used to obtain the position, speed and direction information of the electric vehicle rider and the electric vehicle; The wireless communication module is used to transmit the acquired information to the upper control system and receive the warning information from the upper control system; The voice module is arranged in the ear corresponding area of ​​the safety helmet and is used to play the warning information; The control module is used to connect and control the camera, microwave sensor, positioning module, wireless communication module and voice module.

[0018] A system for warning of ghost-heading traffic accidents, comprising a host control system, multiple positioning modules, a gateway and the above-mentioned safety helmet, wherein the safety helmet is worn by each electric vehicle rider during riding, and is used to obtain the driving environment, vehicle distance, vehicle flow, coordinate position and speed direction of the corresponding electric vehicle rider, and play warning information; A positioning module, used to obtain the position, speed and direction of the corresponding motor vehicle; Gateway, used to transmit the data obtained by the safety helmet and positioning module to the upper control system; The upper control system is used to use the SAN-GAN algorithm to obtain an interaction sequence based on the data obtained from the electric vehicle rider, use the pre-trained neural network model to generate hidden feature information of the electric vehicle rider according to the interaction sequence, and predict the future trajectory of the electric vehicle rider according to the hidden feature information; based on the predicted future trajectory, use the uncertainty position prediction method, introduce a proportional factor, and calculate and predict the collision area between the electric vehicle rider and the dangerous vehicle; based on the collision area, judge the longitudinal and lateral collision danger state of the electric vehicle rider according to the collision danger state evaluation index, obtain the danger state judgment result, form early warning information, and send it to the corresponding safety helmet.

[0019] Compared with the prior art, the present invention has the following beneficial effects: (1) Different from the research on early warning from the perspective of automobiles, the present invention uses V2P communication, GPS and other technologies, combined with LSTM neural network, collision risk status evaluation index and other algorithms to design an early warning helmet that integrates road environment, location information collection, real-time perception, trajectory prediction, accident warning, and real-time voice communication. The receiver of the early warning information is converted to the electric vehicle rider, so that the electric vehicle rider can actively prevent the occurrence of collision accidents.

[0020] (2) The trajectory prediction using the SAN-GAN algorithm in the present invention is more reasonable than the collision research that only considers the linear trajectory prediction of the rider. The four collision hazard state evaluation indicators for judging the collision risk effectively reduce the false alarm rate and the false alarm rate of the early warning compared with the traditional collision time algorithm TTC, and can more effectively protect the safety of electric vehicle riders.

[0021] (3) The present invention has a wide range of practical application scenarios and can also be used in other sections with complex traffic. The helmet can be connected to the enterprise systems of traffic control and express delivery to expand the scope of applicable groups, further improve the applicability and information interaction capabilities of the helmet, provide more real-time road information for riders, and improve the accident warning capabilities of the helmet. It is of great significance to reduce the incidence of urban ghost-heading traffic accidents, solve urban traffic safety problems, promote the high-end, intelligent, and digital high-quality development of the transportation infrastructure system, and promote the intelligent innovation and development of transportation equipment technology.

[0022] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0024] Figure 1 This is an architecture diagram of a ghost-heading traffic accident warning system according to an embodiment; Figure 2 A design flow chart of a ghost head traffic accident warning system according to an embodiment; Figure 3 A flow chart of an electric vehicle rider trajectory prediction embodiment; Figure 4 A schematic diagram of the interaction of the visual field area of ​​an electric vehicle rider according to an embodiment; Figure 5 A schematic diagram of a collision area ellipse according to an embodiment; Figure 6 The position relationship between a dangerous vehicle, an obstructing vehicle and an electric vehicle rider in one embodiment; Figure 7 A schematic diagram of coordinates of a dangerous vehicle according to an embodiment; Figure 8 A schematic diagram of a lateral collision hazard according to an embodiment; Fig. 9 1 is a simulation experiment result diagram of an embodiment, wherein (a) is the false alarm rate and (b) is the false alarm rate. DETAILED DESCRIPTION

[0025] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0026] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0027] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0028] In the absence of conflict, the embodiments in this application and the features in the embodiments may be combined with each other.

[0029] Embodiment 1 A safety helmet for early warning of ghost-heading traffic accidents, comprising a safety helmet body, on which a camera, a microwave sensor, a positioning module, a wireless communication module, a voice module and a control module are arranged, wherein: The camera is arranged in front of the safety helmet body and is used to obtain dynamic image information of the road surface. In this embodiment, a micro camera is selected; The microwave sensor is arranged in front of the safety helmet body and is used to obtain vehicle flow, instantaneous vehicle speed and vehicle distance information, that is, environmental information; The positioning module is used to obtain the position, speed, direction and coordinate information of the electric vehicle rider and the electric vehicle. In this embodiment, a GPS module is used; The wireless communication module is used to transmit the acquired information to the upper control system and receive the warning information of the upper control system. In this embodiment, the LoRa module is selected; The voice module is arranged in the ear corresponding area of ​​the safety helmet and is used to play the warning information; The control module, in this embodiment, uses an STM32 chip to connect and control the camera, microwave sensor, positioning module, wireless communication module and voice module.

[0030] Embodiment 2 A system for warning of ghost-like traffic accidents, such as Figure 1 As shown, it includes a host control system, multiple positioning modules, a gateway and a safety helmet of the first embodiment, wherein the safety helmet is worn by each electric vehicle rider during riding, and is used to obtain the driving environment, vehicle distance, vehicle flow, coordinate position and speed direction of the corresponding electric vehicle rider, and play warning information; A positioning module is used to obtain the position, speed and direction of motor vehicles within the corresponding range. In this embodiment, a vehicle terminal GPS is used; This range is mainly because the traffic control center system will identify the motor vehicles that are close to the electric vehicle, and obtain the relevant position, speed and direction of the dangerous vehicle from the vehicle terminal GPS of the motor vehicle (dangerous vehicle) combined with the microwave sensor on the helmet. That is, the relevant position, speed and direction information of the electric vehicle and the motor vehicles around it are transmitted to the traffic control center system through the wireless communication network.

[0031] Gateway, used to transmit the data obtained by the safety helmet and positioning module to the upper control system; The upper control system, such as Figure 1As shown, it is used to use the SAN-GAN algorithm to obtain an interaction sequence based on the data obtained from the electric vehicle rider, use the pre-trained neural network model to generate hidden feature information of the electric vehicle rider according to the interaction sequence, and predict the future trajectory of the electric vehicle rider according to the hidden feature information; based on the predicted future trajectory, use the uncertainty position prediction method, introduce a proportional factor, and calculate and predict the collision area between the electric vehicle rider and the dangerous vehicle; based on the collision area, judge the longitudinal and lateral collision danger state of the electric vehicle rider according to the collision danger state evaluation index, obtain the danger state judgment result, form early warning information, and send it to the corresponding safety helmet.

[0032] The process of judging the danger state of longitudinal and lateral collision includes: Establishing the time remaining for collision The vertical dangerous state judgment model is: ; In the formula, are the relative speed and relative acceleration of the dangerous vehicle and the electric bike rider respectively; D is the longitudinal distance between the dangerous vehicle and the electric bike rider at the collision position; L is the body length of the dangerous vehicle. is the shortest time for an electric bike rider to reach the dangerous collision area, predicted by the LSTM neural network.

[0033] Combine the collision risk area obtained by trajectory prediction and calculate and compare in real time and The size of the electric bike is used to determine whether the electric bike rider is in a dangerous state in the longitudinal direction: > When , it is in safe state, and the output longitudinal danger signal value is 0; when ≤ , it is a longitudinal collision danger state, and the longitudinal danger signal value 1 is output.

[0034] The time when an electric bike rider enters and leaves the dangerous collision zone , for: ; ; In the formula, is the width of the dangerous vehicle; is the width of the electric bike rider; is the minimum safe distance that the dangerous vehicle and the electric bike rider need to maintain; S is the lateral distance at the collision position between the dangerous vehicle and the electric bike rider; is the lateral speed of the electric bike rider relative to the coordinate system of the dangerous vehicle.

[0035] Analysis of lateral collision conditions: According to the calculation , compare the size relationship between the three: a. When , it is in safe state, and the lateral danger signal value is 0; b. When , it is in safe state, and the lateral danger signal value is 0; c. , it is a lateral danger state and the lateral danger signal value 1 is output.

[0036] If both the longitudinal and lateral danger states output a danger signal value of 1, it is judged that a collision risk will occur. The traffic control center transmits the danger signal to the helmet terminal through the gateway, and the helmet issues warning information to the electric vehicle rider through the voice module.

[0037] Host control system, such as Figure 1 As shown, it is set in the traffic control center and can be a cloud server. The cloud server is connected to an edge computing unit, and each edge computing unit runs in parallel to improve the computing and processing efficiency.

[0038] The cloud server obtains the positioning information of the positioning module of the vehicle terminal and the multi-dimensional perception information of the safety helmet terminal through the gateway. The safety helmet terminal transmits the acquired data to the gateway through the LoRa module.

[0039] like Figure 2 As shown, each edge computing unit processes the traffic accident warning process of a certain area: obtaining the position, speed and direction of each motor vehicle within a certain range of the current driving position of the electric vehicle rider; Using the SAN-GAN algorithm, based on the data obtained from the electric bike rider, the interaction sequence is obtained. Using the pre-trained neural network model, the hidden feature information of the electric bike rider is generated according to the interaction sequence, and the future trajectory of the electric bike rider is predicted based on the hidden feature information. Based on the predicted future trajectory, the uncertainty position prediction method is used to introduce a proportional factor to calculate and predict the collision area between the electric bike rider and the dangerous vehicle; Based on the collision area, the longitudinal and lateral collision danger state of the electric vehicle rider is judged according to the collision danger state evaluation index, the danger state judgment result is obtained, and early warning information is formed.

[0040] A ghost collision accident is an accident in which a motor vehicle (dangerous vehicle) and a non-motor vehicle (electric vehicle rider) appear from outside the sight triangle of each other at a close distance and for a short time due to the influence of a stopped obstructing vehicle, leaving neither party with time and space to react or evade, and eventually a collision occurs. Therefore, vehicles that are at risk of ghost collision with electric vehicles are defined as dangerous vehicles.

[0041] The cloud server sends the corresponding warning information to the safety helmet terminal, and the rider brakes the electric vehicle.

[0042] In some embodiments, the safety helmet terminal may also have other functions, such as detecting whether the safety helmet is worn incorrectly before riding, whether the rider is speeding on a normal road section, and whether a fall may occur.

[0043] Embodiment 3 A method for early warning of a ghost-heading traffic accident, applying the above embodiment, comprises the following steps: Step 1: Information collection; The helmet terminal transmits the driving environment, vehicle distance, vehicle flow, coordinate position and other information collected by the camera, microwave sensor and GPS to the traffic control center (cloud server) in real time. The typical scenes of ghost peeking on urban roads are divided into three typical categories: (1) Ghost peeking at a signalized intersection At a flat signalized intersection, straight-moving vehicles line up in front of the stop line under a no-passing signal, creating a blind spot for electric bike riders relative to the right-turn lane. The right-turn lane is generally not controlled by traffic lights and is free to pass, so there is a risk of collision between the right-turn lane and the relatively flexible electric bikes when approaching the intersection.

[0044] (2) Urban T-intersection ghost peeking scene Vehicles parked on the roadside at T-shaped intersections create a blind spot between electric vehicle riders and straight-moving motor vehicles. At urban T-shaped intersections, there is a certain risk of collision between electric vehicles and motor vehicles (or simply vehicles).

[0045] (3) Scene of ghosts poking out of the city branch road Urban branch roads are characterized by complex composition, large number of pedestrians, complex types of vehicles, multiple intersections, and dispersed traffic; electric vehicle riders show greater randomness and flexibility on branch roads; and vehicle parking on branch roads is more likely to cause blind spots.

[0046] Step 2: Information processing and trajectory prediction; Specifically include: Step 21, SAN-GAN predicts the trajectory of the electric bike rider; The present invention adopts SAN-GAN algorithm to predict the trajectory of electric bike riders. The historical location information and head information of electric bike riders are collected, and the hidden feature information of electric bike riders is obtained through the LSTM neural network in the trajectory generator. The electric bike rider's field of view function model is established and effective information is screened, thereby driving the neural network model to predict the future trajectory changes of electric bike riders.

[0047] The process of predicting the trajectory of an electric bike rider is as follows: Figure 3 As shown, specifically including: Step 201: collecting and preprocessing the electric vehicle rider's location data; The DJI--MINI2 drone was used for data collection. The pedestrian trajectory prediction method based on the field of view mechanism was used, and CMOS was used as the image sensor. The shooting survey was conducted at the intersection of Youyi East Road and Yanta Road in Xi'an. The shooting scene was a rectangular area with a length of 20m and a width of 12m. The trajectory of the electric bike rider and the deflection angle of the rider's head were calibrated every 10 frames in the video, and the two-dimensional coordinates of the i-th electric bike rider were defined ( , ) and its head deflection angle , 352 valid electric bike rider trajectories and head deflection angles were sorted out. The calibration data was used as the training set and test set for the algorithm experiment. 80% was selected as the training set and validation set, and 20% was selected as the test set. After the collected data was processed by the SAN-GAN algorithm, the electric bike rider interaction sequence { , , …, }, ={( , ), Taking the data collection of the 10th rider as an example, the partial trajectory and head angle information table is shown in Table 1: Table 1 Part of the calibration data

[0048] The training data is input into the SAN-GAN algorithm, and it is iterated in the trajectory generator to form the ability to quickly memorize and process data. In real-time conditions, the algorithm can continuously and quickly and accurately calculate the position data from the perception layer to obtain the predicted trajectory.

[0049] Step 202: Obtain hidden feature information of the electric vehicle rider; The LSTM neural network embeds the prior information into the training data as the hidden features of the electric bike rider’s trajectory. The trajectory generator is used to generate the hidden feature information of the electric bike rider, which includes a trajectory encoder, an electric bike rider’s field of view pooling module, and a trajectory decoder. The encoder module encodes the feature vector of the electric bike rider i , the historical trajectory feature vector of the electric bike rider is obtained through LSTM neural network encoding : (1) (2) In the formula, FC is the fully connected neural network layer; is the weight parameter in FC; is the weight parameter in the LSTM neural network layer.

[0050] Step 203, establishing a visual field function model; It also includes: (1) Establish the visual field of electric vehicle riders; Consider the head turning angle of the electric bike rider based on the original electric bike rider position relationship Based on the factors of the interaction between electric vehicle riders, the electric vehicle rider’s field of view is established. The electric vehicle rider’s field of view is the sector area, such as Figure 4 The following is a schematic diagram of the interaction of the electric vehicle rider's field of vision: Combined with the actual change of the electric bike rider's field of view during the interaction process, the electric bike rider's field of view is established. : (3) (4) (5) In the formula, is the visual field of the i-th electric bike rider at time t; for The polar diameter; l is the length of the angle bisector that bisects the area of ​​the field of view; For this moment the polar angle of the electric bike rider; is the head deflection angle of the electric bike rider at this moment, that is l Angle with the horizontal axis; The actual field of view of the electric bike rider. is the relative distance between electric bike riders i and j.

[0051] (2) Establish the visual field function of the electric vehicle rider in the interactive state; Considering that multiple other electric bike riders are in the field of vision of the predicted electric bike rider at the same time, the predicted electric bike rider is taken as the influence source. The closer the electric bike rider in the field of vision is to the influence source, the greater the repulsive force it generates, and the farther the distance is, the smaller the repulsive force generated. Based on this, the field of vision function of the electric bike rider is established. , and obtain the impact of the position information on the electric bike rider : (6) (7) In the formula, is the repulsion amplitude parameter between electric bike riders; r is the minimum collision-free distance between electric bike riders; represents the visual range of electric bike rider i at time t; is the field of view function of the electric bike rider; for The judgment parameters in .

[0052] Step 204, determining the predicted trajectory of the electric vehicle rider; Through the visual field module function in formulas (6) and (7), the hidden feature vector of the electric bike rider’s visual field is output: , and then the relative distance between electric bike rider i and other electric bike riders is output through the fully connected layer to obtain the feature vector , the two feature vectors are combined and processed to obtain the field of view pooling vector After the trajectory decoder inputs data into the LSTM neural network layer and multi-layer perception, the trajectory prediction point of the electric vehicle rider can be obtained ( , ). (2) Experimental analysis of SAN-GAN algorithm: To verify the effectiveness of SAN-GAN in predicting the trajectory of electric bike riders, LSTM, S-LSTM, S-GAN and SAN-GAN were used to conduct the same experimental comparison in Pytroch-Adam, and ADE and FDE were used as evaluation indicators. ADE is represented by the average distance between the predicted trajectory and the historical true trajectory of all tested electric bike riders at each moment. FDE represents the final displacement distance between the predicted trajectory and the historical true trajectory of all tested electric bike riders at the last moment.

[0053] The input observation duration is 8 frames, and the output prediction duration is This is a comparative experiment of 8 frames (3.2s) and 12 frames (4.8s). The experimental results show that as the number of iterative training increases, the trend of loss decreases gradually slows down, and the evaluation index eventually converges. The error values ​​of the four algorithms are extracted for comparative analysis, as shown in Table 2: Table 2 Comparison of trajectory prediction algorithm errors

[0054] It can be seen that the SAN-GAN algorithm comprehensively considers the influence of the geographical location and environmental factors of electric vehicle riders, and the predicted trajectory is closest to the actual trajectory, which improves the interpretability and accuracy of the electric vehicle rider trajectory prediction.

[0055] Step 3: Collision area analysis: In existing studies, the maximum likelihood solution or the difference of the major axis of the ellipse is mostly used to describe the collision area. The future uncertain trajectory of the electric bike rider is represented as the collision area given by the confidence ellipse through the LSTM neural network.

[0056] Step 31, constructing a collision area; Through LSTM, we can get the mean and covariance of the future trajectory of the electric bike rider. Assume that the probability of being at a certain position depends on the collision area Ri, which is determined based on the trajectory mean and covariance matrix at time t. Use the square root of the covariance matrix to map a unit circle and determine an ellipse whose major semi-axis depends on the eigenvalue of the matrix and its direction is related to the corresponding eigenvector.

[0057] Step 32, calculating the confidence probability; By using the inverse Mahalanobis transform and introducing different scaling factors, the confidence probability of any ellipse can be calculated. When the scaling factor r is 2.7955 and 3.3682, the corresponding confidence probability P(r) is 95% and 99% respectively. The electric bike rider may move in a straight line or in a non-straight line. In both cases, 6 seconds or 10 seconds are used to predict the probability of a position offset of 0 meters. By inputting different data, the confidence probability of the collision area is obtained, as shown in Table 3: Table 3 Confidence probability results

[0058] The first number 10 is the number of data used by LSTM, and the second number 2 is the position data predicted by LSTM after 2 seconds. 6.5.1 means that 6 data are used to predict the position of the electric bike rider in 5 seconds.

[0059] The LSTM neural network can be used to obtain the mean and covariance of the trajectory of the electric bike rider and then predict the trajectory. The experimental results show that the probability density function of the trajectory of the electric bike rider obeys the multidimensional normal distribution function, which infers that the collision area is not a single point. By using the uncertainty position prediction method and introducing the proportional factor, it is concluded that the collision area can be represented by a confidence ellipse, such as Figure 5 As shown in the figure, the horizontal coordinates are the latitude and longitude after the vertical coordinate system is mapped to the plane coordinate system. The center point of the ellipse is the average value, and the colored area is the collision area. The closer to the center point, the smaller the covariance, and the higher the probability of collision.

[0060] Step 4: Determine the risk of dangerous collision; Based on trajectory prediction, the collision area between the electric vehicle rider and the dangerous vehicle is calculated. The traffic control center (cloud server) further judges the electric vehicle rider's longitudinal and lateral collision risk behavior through the collision risk status evaluation index, and sends the risk behavior judgment result, i.e. warning information, to the helmet terminal at the application layer.

[0061] For the anti-collision warning system, the safe time algorithm can better reflect the perception and judgment habits of dangerous vehicle drivers and electric vehicle riders in natural driving and riding states compared to the safe distance algorithm.

[0062] Based on the safety time difference algorithm and combined with specific scenarios, four collision danger state evaluation indicators are established according to the movement state and position relationship between the dangerous vehicle and the electric bike rider to realize the judgment of the longitudinal and lateral danger state of the electric bike rider: a. The time it takes for the dangerous vehicle to reach the dangerous collision area in the longitudinal direction (Time To Collision, TTC); b. The time it takes for the electric bike rider to reach the dangerous collision area in the longitudinal direction (Time To People, TTP); c. The time it takes for the electric bike rider to enter the dangerous collision area in the lateral direction relative to the dangerous vehicle (Time To Enter, TTE); d. The time it takes for the electric bike rider to leave the dangerous collision area in the lateral direction relative to the dangerous vehicle (Time To Leave, TTL).

[0063] Step 41: Build Longitudinal dangerous state judgment model; The helmet terminal GPS obtains the movement status information and coordinate location information of the electric bike rider, and the vehicle terminal GPS obtains the movement status information and coordinate location information of the dangerous vehicle, and combines the prediction of the electric bike rider's trajectory, taking into account the remaining time of the collision (that is, the time it takes for the car to maintain the current speed to reach the collision point) to establish a collision prediction algorithm. The longitudinal risk state judgment model is composed of Figure 6 The equation of motion is: (8) In the formula, , are the speeds of dangerous vehicles and e-bike riders, respectively; , are the accelerations of dangerous vehicles and electric bike riders, respectively; is the relative distance between the center of the dangerous vehicle and the center of the electric bike rider; For the vehicle commander; is the distance from the center of the dangerous vehicle to the front bumper mask of the dangerous vehicle; It is the distance from the electric bike rider to the rear bumper mask.

[0064] This embodiment takes the dangerous collision scenario between an electric bike riding on a zebra crossing at an urban road intersection and a vehicle turning right without signal control on the right turn lane as an example for analysis. The motion state information and coordinate position information of the electric bike rider and the dangerous vehicle are obtained according to the GPS of the helmet terminal and the vehicle terminal, and a longitudinal dangerous state judgment model is established to judge the dangerous state in combination with the prediction of the electric bike rider's trajectory.

[0065] Establishing the time remaining for collision The vertical dangerous state judgment model is: ; In the formula, are the relative speed and relative acceleration of the dangerous vehicle and the electric bike rider respectively; D is the longitudinal distance between the dangerous vehicle and the electric bike rider at the collision position; L is the body length of the dangerous vehicle. is the shortest time for an electric bike rider to reach the dangerous collision area, predicted by the LSTM neural network.

[0066] Combine the collision risk area obtained by trajectory prediction and calculate and compare in real time and The size of the electric bike is used to determine whether the electric bike rider is in a dangerous state in the longitudinal direction: > When , it is in safe state, and the output longitudinal danger signal value is 0; when ≤ , it is a longitudinal collision danger state, and the longitudinal danger signal value 1 is output.

[0067] Step 42: Establishing a global coordinate system and a vehicle coordinate system; When an electric bike rider passes in front of an obstructing vehicle, the distance, angle and other information between the rider and the obstructing vehicle can be obtained by GPS data and calculated by the traffic control center (cloud server). Assume that the center of the front bumper mask of the obstructing vehicle is the origin of the coordinate system with the obstructing vehicle as the coordinate system. Figure 7 As shown, the global coordinate system is XOY, and the vehicle coordinate system is X′O′Y′. To facilitate coordinate conversion, the local coordinate system after translation of the global coordinate system XOY is set to X″O″Y ″, (X1, Y1) is the coordinate of the blocking vehicle in the global coordinate system, and (X2, Y2) is the coordinate of the electric vehicle rider in the global coordinate system.

[0068] Step 43: Convert the electric bike rider's position coordinate system; analyze Figure 7 The coordinates of the electric bike rider in the vehicle coordinate system of the blocking vehicle can be obtained (X′2, Y′2), and the horizontal and vertical coordinates are: (9) In the formula, is the distance from the center of the vehicle to the center of the front end of the vehicle; R is the straight-line distance between the vehicle and the rider; is the azimuth angle of the electric bike rider in front relative to the vehicle.

[0069] According to the different ranges of the heading angle of the blocking vehicle, the analysis can be divided into the following two types: When the heading angle of the blocking vehicle When the coordinates of the electric bike rider in the vehicle coordinate system of the blocking vehicle are and the coordinates of the occluding vehicle in the global coordinate system , we can get the coordinates of the electric bike rider in the global coordinate system for: (10) when When , the coordinates of the electric bike rider in the global coordinate system can be expressed as: (11) Step 44: Calculate the relative position between the dangerous vehicle and the electric bike rider; According to the coordinates of the electric bike rider in the global coordinate system and the coordinates of the dangerous vehicle in the global coordinate system , we can know the relative position relationship between dangerous vehicles and electric vehicle riders, such as Figure 6 As shown, the relative distance between dangerous vehicles and electric bike riders The angles β between the line connecting the two and the y-axis are: (12) (13) The longitudinal distance D and lateral distance S of the collision position between the dangerous vehicle and the electric bike rider and the direction angle of the electric bike rider They are: (14) (15) (16) In the formula, are the heading angles of the dangerous vehicle and the blocking vehicle, that is, the angle between the vehicle's driving direction and the y-axis (due north); are the longitudinal and lateral velocities of the electric bike rider in the coordinate system with the blocking vehicle as the coordinate.

[0070] The speed of the electric bike rider Transform to obtain the longitudinal speed of the electric bike rider relative to the vehicle coordinate system of the dangerous vehicle and lateral speed : (17) Step 45: Standardize the values ​​of some parameters of the lateral collision risk judgment model; Establish a lateral collision risk judgment model, such as Figure 8 shown.

[0071] In this embodiment, in addition to determining the longitudinal danger state, it is also necessary to consider whether there is a potential collision risk between the dangerous vehicle and the electric bike rider in the lateral direction. The TTE and TTL evaluation indicators are established based on the motion state and position relationship between the dangerous vehicle and the electric bike rider, and the TTC indicator is combined to determine whether there is a collision risk for the electric bike rider in the lateral direction.

[0072] The time when an electric bike rider enters and leaves the dangerous collision zone , for: ; ; In the formula, is the width of the dangerous vehicle; is the width of the electric bike rider; is the minimum safe distance that the dangerous vehicle and the electric bike rider need to maintain; S is the lateral distance at the collision position between the dangerous vehicle and the electric bike rider; is the lateral speed of the electric bike rider relative to the coordinate system of the dangerous vehicle.

[0073] Analysis of lateral collision conditions: According to the calculation , compare the size relationship between the three: a. When , it is in safe state, and the lateral danger signal value is 0; b. When , it is in safe state, and the lateral danger signal value is 0; c. , it is a lateral danger state and the lateral danger signal value 1 is output.

[0074] If both the longitudinal and lateral danger states output a danger signal value of 1, it is judged that a collision risk will occur. The traffic control center transmits the danger signal to the helmet terminal through the gateway, and the helmet issues warning information to the electric vehicle rider through the voice module.

[0075] After consulting a lot of information, we know that cars are divided into small cars, medium-sized cars and large cars according to their wheelbases. The body width is in the range of 1600-2000mm. We take medium-sized cars as an indicator and take 1800mm as the width of dangerous vehicles. 1800mm; According to the quality requirements for two-wheeled electric vehicles in the "Electric Bicycle Safety Technical Specifications" standard, the center distance between the front and rear wheels of an electric bicycle shall not exceed 1.25m, and the width of the vehicle body shall not exceed 0.45m. The width for e-bike riders is 1.25m; According to the relevant provisions of the "Safe and Civilized Operation Specifications for Motor Vehicle Drivers", the longitudinal distance between vehicles and pedestrians in the same direction of travel must be greater than 1.5m. That is, the minimum safe distance that dangerous vehicles need to maintain from electric vehicle riders is 1.5m.

[0076] Step 46: Setting up a simulation experiment environment for collision risk state judgment; The selected simulation scene is the signalized intersection in the above three scenes. The type of dangerous vehicle is the sedan model built into the simulation software. The main variables of the experiment are the speed of dangerous vehicles and electric bike riders passing through the crosswalk area of ​​the intersection. The dangerous vehicle is 50 m away from the zebra crossing, and the electric bike rider is 1 m away from the zebra crossing. The speed of the electric bike rider is divided into low speed, medium speed and high speed, which are (0, 1] m / s, (1, 5] m / s and (5, 10] m / s respectively. The speed of dangerous vehicles is divided into low speed, medium speed and high speed, which are (10, 30] km / h, (30, 60] km / h and (60, 90] km / h respectively. According to the three speed levels of electric bike riders and dangerous vehicles, a total of 9 groups of experiments were conducted, see Table 4.

[0077] Table 4 Simulation experiment parameters

[0078] Specifically include: 1) Construction of intersection simulation scene; The simulation experiment platform uses VISSIM software, and the experiment selects PTV VISSIM 8 (student) version to test the dangerous collision time difference model. A traffic simulation environment is established, and a dangerous scenario is built in which an electric bike rider passes in front of a stationary blocking vehicle and collides with a dangerous vehicle. This scenario is used to compare and verify the four dangerous state collision time difference indicators TTC, TTP, TTE, and TTL proposed in this embodiment with the traditional TTC.

[0079] According to the software's own model, dangerous vehicles are set to travel slowly at the speed set at the beginning and fluctuate within a certain reasonable range; electric bike riders plan their paths through the pre-set starting point and the end point obtained after trajectory prediction. Output the motion posture information of dangerous vehicles and electric bike riders during the entire simulation process. This simulation set a total of 9 groups of control experiments, in which the speeds of dangerous vehicles were set to 30 km / h, 60 km / h and 90km / h, and the speeds of electric bike riders were set to 1.0 m / s, 5.0 m / s and 10.0 m / s.

[0080] 2) Simulation experiment results and analysis; In the process of statistical early warning experiment, the experiment was divided into 9 groups according to the speed of dangerous vehicles and electric bike riders. Each group of experiments was conducted 100 times, and the simulation time was 30 seconds. Statistics were obtained under different experimental conditions to compare the four collision danger state evaluation indicators of TTC, TTP, TTE, and TTL proposed in this embodiment with the traditional TTC. The simulation experiment mainly uses the false alarm rate and the false alarm rate of the experiment to comprehensively evaluate the collated experimental data. The experimental results are as follows: Fig. 9 As shown, (a) is a comparison chart of the false alarm rates of different algorithm warnings, and (b) is a comparison chart of the false alarm rates of different algorithm warnings. This warning is the warning result of this embodiment.

[0081] False alarm rate analysis False alarm means that the electric bike rider was warned too early, resulting in ineffective collision avoidance, that is, at the time of warning the electric bike rider, the dangerous vehicle was still relatively far away from the intersection pedestrian conflict area. False alarm rate It can be expressed as: (18) Where: n1 is the number of times the visual warning is given too early to the electric vehicle rider; N is the total number of simulation experiments.

[0082] False alarm rate analysis False alarm refers to the warning that a traffic accident still occurs due to the late warning of electric bike riders. That is, in the simulation experiment of vehicles and electric bike riders in the pedestrian area of ​​the intersection, it is believed that although the electric bike riders are warned, a traffic accident still occurs. It can be expressed as: (19) Where n2 is the number of times an accident still occurred due to a late visual warning for electric vehicle riders.

[0083] Step 5: Release of early warning information; The helmet terminal transmits the warning information to the electric vehicle rider through the voice module.

[0084] As the system's perception layer terminal, the helmet has multi-dimensional perception functions and is composed of a single-chip STM32 chip, a LoRa module, a micro camera, a microwave sensor, a GPS with sub-meter accuracy, a power amplifier module, and a voice module. The camera collects dynamic image information of the driving environment; the microwave sensor collects information such as vehicle flow, instantaneous vehicle speed, and vehicle distance; GPS technology obtains real-time information such as the location, speed, and direction of electric bike riders and motor vehicles.

[0085] The safety helmet terminal transmits the acquired data to the gateway through the LoRa module. The gateway uploads all uplink data packets to the traffic control center system. The traffic control center includes four major functional modules: (1) Comprehensive situation display: display various functional indicators of the system; (2) Real-time traffic analysis: real-time display of the cyclist’s driving environment information; (3) Early warning monitoring: the analysis process that displays the collision risk of the road environment; (4) Warning status analysis: Displays the process and results of the system's judgment and confirmation of the monitored risks. The traffic control center obtains warning information after analysis and calculation, and finally transmits the warning information to the helmet terminal, which releases the warning information to the electric vehicle rider through the voice module.

[0086] It should be understood by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of an all-hardware embodiment, an all-software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of one or more computer-usable storage media (including but not limited to disk storage, CD - ROM , optical storage, etc.).

[0087] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0088] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0089] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0090] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principle of the present invention without creative labor shall be included in the protection scope of the present invention.

Claims

1. A method for early warning of ghost-probing traffic accidents, characterized in that: The following steps are involved: Obtain the driving environment, vehicle distance, traffic flow, coordinate position and speed direction of the electric vehicle rider; Obtain the position, speed and direction of each motor vehicle within a certain range of the electric vehicle rider's current driving position; Using the SAN-GAN algorithm, based on the data obtained from the electric bike rider, the interaction sequence is obtained. Using the pre-trained neural network model, the hidden feature information of the electric bike rider is generated according to the interaction sequence, and the future trajectory of the electric bike rider is predicted based on the hidden feature information. Based on the predicted future trajectory, the uncertainty position prediction method is used to introduce a proportional factor to calculate and predict the collision area between the electric bike rider and the dangerous vehicle; Based on the collision area, the longitudinal and lateral collision danger state of the electric vehicle rider is judged according to the collision danger state evaluation index, the danger state judgment result is obtained, and early warning information is formed.

2. A method for early warning of ghost-heading traffic accidents as claimed in claim 1, characterized in that: The training process of the neural network model includes: collecting images in the target area, obtaining the trajectory of each electric bike rider and the deflection angle of the rider's head at regular intervals, and calibrating them to define the two-dimensional coordinates of the i-th electric bike rider. Head angle , multiple sets of effective electric vehicle rider trajectories and head deflection angles are obtained, at least a part of them are taken as training data, and after being processed by the SAN-GAN algorithm, the interaction sequence of each electric vehicle rider is obtained to train the neural network model.

3. The method for early warning of ghost-heading traffic accidents as claimed in claim 1, characterized in that: The process of generating hidden feature information of an electric vehicle rider according to an interaction sequence includes: using a trajectory generator to generate hidden feature information of an electric vehicle rider, the trajectory generator comprising a trajectory encoder, an electric vehicle rider field pooling module and a trajectory decoder, the trajectory encoder being used to encode to obtain a feature vector of a target electric vehicle rider, and obtain a feature vector of a historical trajectory of the electric vehicle rider; The electric vehicle rider field of view pooling module is used to establish the electric vehicle rider field of view based on the interaction relationship between the electric vehicle riders, establish the electric vehicle rider field of view function, and obtain the electric vehicle rider field of view hidden feature vector; The relative distance between the target electric bike rider and other electric bike riders is passed through a fully connected layer to obtain a feature vector, which is combined with the hidden feature vector of the electric bike rider’s field of view and used as the input of the trajectory decoder; The trajectory decoder is used for decoding to obtain the trajectory prediction point of the electric vehicle rider.

4. A method for early warning of ghost-heading traffic accidents as claimed in claim 3, characterized in that: The process of establishing the electric vehicle rider's field of vision based on the interaction relationship between electric vehicle riders includes: combining the actual change state of the electric vehicle rider's field of vision during the interaction process, establishing the electric vehicle rider's field of vision : ； ； ； in, is the visual field of the i-th electric bike rider at time t; for The polar diameter; l is the length of the angle bisector that bisects the area of ​​the field of view; For this moment the polar angle of the electric bike rider; is the head deflection angle of the electric bike rider at this moment, that is l Angle with the horizontal axis; The actual field of view of the electric bike rider. is the relative distance between electric bike riders i and j.

5. A method for early warning of ghost-heading traffic accidents as claimed in claim 3, characterized in that: The electric bike rider's field of view function, the process of obtaining the electric bike rider's field of view hidden feature vector includes: considering that multiple other electric bike riders are in the field of view of the predicted electric bike rider at the same time, taking the predicted electric bike rider as the influence source, the closer the electric bike rider in the field of view is to the influence source, the greater the repulsive force it generates, and the farther the distance is, the smaller the repulsive force generated, and thus establishing the electric bike rider's field of view function , and obtain the impact of the position information on the electric bike rider : ； ； In the formula, is the repulsion amplitude parameter between electric bike riders; r is the minimum collision-free distance between electric bike riders; represents the visual range of electric bike rider i at time t; is the field of view function of the electric bike rider; for The judgment parameters in .

6. The method for early warning of ghost-heading traffic accidents as claimed in claim 1, characterized in that: The process of calculating and predicting the collision area between an electric bike rider and a dangerous vehicle includes: using the neural network model to obtain the mean and covariance of the future trajectory of the electric bike rider, assuming that the probability of the electric bike rider being located at a certain position depends on the collision area Ri, Ri is determined based on the trajectory mean and covariance matrix at time t, and a unit circle is mapped using the square root of the covariance matrix to determine an ellipse, whose major semi-axis depends on the eigenvalue of the matrix, and whose direction is related to the corresponding eigenvector; using the inverse Mahalanobis transform, introducing different proportional factors to calculate the confidence probability of any ellipse, the area of ​​the collision area is represented by the confidence ellipse, and the closer to the center position of the ellipse, the higher the collision probability.

7. The method for early warning of ghost-heading traffic accidents as claimed in claim 1, characterized in that: The collision danger status evaluation index includes the time when the dangerous vehicle reaches the dangerous collision area in the longitudinal direction, the time when the electric vehicle rider reaches the dangerous collision area in the longitudinal direction, the time when the electric vehicle rider enters the dangerous collision area in the lateral direction relative to the dangerous vehicle, and the time when the electric vehicle rider leaves the dangerous collision area in the lateral direction relative to the dangerous vehicle.

8. The method for early warning of ghost-heading traffic accidents as claimed in claim 1, characterized in that: Before making a judgment on the dangerous state of longitudinal and lateral collision, obtain the distance and angle between the electric bike rider and the blocking vehicle, construct a coordinate system with the blocking vehicle as the coordinate origin of the coordinate system, and transform the electric bike rider's position coordinate system to determine the coordinates of the electric bike rider in the global coordinate system. According to the coordinates of the electric bike rider in the global coordinate system and the coordinates of the dangerous vehicle in the global coordinate system, determine the relative position relationship between the dangerous vehicle and the electric bike rider, and then calculate the longitudinal speed and lateral speed of the electric bike rider relative to the vehicle coordinate system of the dangerous vehicle.

9. A method for early warning of ghost-heading traffic accidents as claimed in claim 1, 7 or 8, characterized in that: The process of determining the longitudinal collision danger state includes: establishing a longitudinal collision danger state based on the obtained motion state information and coordinate position information of the electric vehicle rider, as well as the motion state information and coordinate position information of the dangerous vehicle, combined with the prediction of the trajectory of the electric vehicle rider, and taking into account the time it takes for the car to maintain the current speed to reach the collision point. Longitudinal dangerous state judgment model: ； In the formula, are the relative speed and relative acceleration of the dangerous vehicle and the electric bike rider respectively; D is the longitudinal distance between the dangerous vehicle and the electric bike rider at the collision position; L is the length of the dangerous vehicle; is the shortest time for an electric bike rider to reach the dangerous collision area, predicted by the neural network model; Compare and The size of > It is a safe state when the vehicle is in the forward collision state, otherwise, it is a dangerous state of longitudinal collision.

10. A method for early warning of ghost-heading traffic accidents as claimed in claim 1 or 7, characterized in that: The process of judging the lateral collision danger state includes: the time when the electric bike rider enters and leaves the dangerous collision area , for: ； ； In the formula, is the width of the dangerous vehicle; is the width of the electric bike rider; is the minimum safe distance that the dangerous vehicle and the electric bike rider need to maintain; S is the lateral distance at the collision position between the dangerous vehicle and the electric bike rider; is the lateral speed of the electric bike rider relative to the coordinate system of the dangerous vehicle; According to the calculation , compare the size relationship between the three, when When , it is safe; when When , it is safe; when It is a lateral dangerous state.

11. The method for early warning of ghost-heading traffic accidents as claimed in claim 1, characterized in that: The process of obtaining the dangerous state judgment result and forming the warning information includes: if both the longitudinal and lateral dangerous states are dangerous states, it is judged that there will be a collision risk and a warning is issued.

12. A safety helmet for early warning of ghost-looking traffic accidents, characterized in that: It includes a safety helmet body, on which a camera, a microwave sensor, a positioning module, a wireless communication module, a voice module and a control module are arranged, wherein: The camera is arranged in front of the safety helmet body and is used to obtain dynamic image information of the road surface; The microwave sensor is arranged in front of the safety helmet body and is used to obtain vehicle flow, instantaneous vehicle speed and vehicle distance information; The positioning module is used to obtain the position, speed and direction information of the electric vehicle rider and the electric vehicle; The wireless communication module is used to transmit the acquired information to the upper control system and receive the warning information from the upper control system; The voice module is arranged in the ear corresponding area of ​​the safety helmet and is used to play the warning information; The control module is used to connect and control the camera, microwave sensor, positioning module, wireless communication module and voice module.

13. A system for warning of ghost-like traffic accidents, characterized in that: It comprises a host control system, multiple positioning modules, a gateway and a safety helmet for warning of ghost-heading traffic accidents as described in claim 12, wherein the safety helmet is worn by each electric bike rider during riding, and is used to obtain the driving environment, vehicle distance, vehicle flow, coordinate position and speed direction of the corresponding electric bike rider, and play warning information; A positioning module, used to obtain the position, speed and direction of the corresponding motor vehicle; Gateway, used to transmit the data obtained by the safety helmet and positioning module to the upper control system; The upper control system is used to use the SAN-GAN algorithm to obtain an interaction sequence based on the data obtained from the electric vehicle rider, use the pre-trained neural network model to generate hidden feature information of the electric vehicle rider according to the interaction sequence, and predict the future trajectory of the electric vehicle rider according to the hidden feature information; based on the predicted future trajectory, use the uncertainty position prediction method, introduce a proportional factor, and calculate and predict the collision area between the electric vehicle rider and the dangerous vehicle; based on the collision area, judge the longitudinal and lateral collision danger state of the electric vehicle rider according to the collision danger state evaluation index, obtain the danger state judgment result, form early warning information, and send it to the corresponding safety helmet.

Images