Research On Collision Probability Prediction Based On Uncertainty Of Electric Two-wheeler Crossing The Street

The 2020 China In-depth Traffic Accident Study(CIDAS)statistical report shows that electric two-wheelers account for 62% of all Vulnerable Road User(VRU)accidents and 6.26%of fatalities in vehicle collisions.As evidenced by the Ministry of Public Security’s "One Belt,One Helmet" safety guarding campaign,electric two-wheeler accidents have become a hot topic of concern for traffic safety in China.However,a large number of collisions have shown that the interaction of human-vehicle-road factors during the "pre-crash interval"(-5～-2s)before the accident is one of the main causes of accidents.As the main means of transportation,the irregularity of electric two-wheeled vehicles has increased the safety risks of road traffic.Therefore,it is of great scientific significance and research value to study the prediction of the collision probability between vehicles and cyclists due to the uncertainty of electric twowheelers’ motion.This paper first establishes a kinematic model that considers the uncertainty of cyclist’s movement across the street.Based on the typical hazardous pre-collision scenario of electric two-wheeled vehicle accidents,300 cases of real motion trajectories of electric two-wheeled vehicles crossing the motorway are collected by UAV to extract the uncertainty feature parameters of the motion trajectories of electric two-wheeled vehicle cyclists.The mathematical equations for the uncertainty of the electric two-wheeler cyclist’s motion were constructed by combining first-order Markov with Gaussian white noise.The results show that the average simulated velocity of the cyclist in the X-direction is 0.003 m/s,while the actual velocity is0.007 m/s,with an absolute error of 0.4%;the average simulated velocity in the Y-direction is2.774 m/s,while the actual velocity is 2.774 m/s,with an absolute error of 0%;and the simulated trajectory distribution and velocity distribution of the cyclist’s uncertainty motion tend to be consistent with the real distribution.Therefore,the cyclist motion model under the uncertainty of electric two-wheeler motion is established effectively and can be used for the study of collision probability prediction model.Secondly,a collision probability model between the vehicle and the cyclist of the electric two-wheeled vehicle is established.A Euclidean distance model considering the relative position of the vehicle and the cyclist of the electric two-wheeler and the accurate indicators(shortest distance and collision time)for collision risk assessment are proposed.The probability prediction model of vehicle collision with electric two-wheelers was obtained by integrating the product.The results show that the Monte Carlo-based collision probability prediction can be used to describe the collision risk between a vehicle and a cyclist,and the average time taken to introduce the Monte Carlo method in this paper is 276.6 ms,which is 5.3 times faster than the average time taken by the same algorithm in previous studies,indicating that the algorithm is somewhat efficient.Finally,the crash probability prediction model is validated based on in-depth accident cases.Two electric two-wheeler depth accidents were screened and the evasive response of the vehicle with different braking strategies of AEB was analysed using PC-Crash software.The effectiveness of the crash probability prediction model was verified by comparing the test results with the probability prediction results.The results show that the prediction model shows a 100% probability of collision when an accident occurs and a 0% probability of collision when an accident is avoided.In addition,the results based on 2 accidents show that: the maximum reduction in crash probability for the AEB braking strategy that avoided the accident in Case 1was 52% on average;compared to 96.3% in Case 2.Deceleration speed a,collision time TTC and sensor detection angle FOV have a significant effect.Of these,FOV has a significant effect and is a prerequisite for determining whether the risk assessment model can perform the collision probability prediction algorithm.