Prediction Method Of Vehicle Collision Based On Deep Learning And Simulation Data

Auto-driving has become a research hotspot in the field of automobiles,and the safety of auto-driving is one of the most critical issues of auto-driving technology.Efficient and accurate collision avoidance early warning ADAS algorithm plays an important role in ensuring the safety of automatic driving vehicles and pedestrians around them.The existing collision avoidance early warning algorithms are based on the traditional safe distance model,and do not have any prediction function.If ADAS collision avoidance algorithm can predict real-time future collision accidents and consequences in the course of driving,thus reducing some reckless driving behavior of drivers,it will greatly reduce the occurrence of traffic accidents.Based on the large data of collision simulation and deep learning method,this paper presents a data-driven predictive anti-collision early warning algorithm,which can directly predict and calculate the damage degree of vehicles and passengers after the collision of two vehicles in advance,so that the driver can make evasive actions in advance after knowing the corresponding dangerous degree,so as to prevent the occurrence of accidents.The main contents of this paper are as follows:(1)Based on PC-Crash,large data of collision simulation for deep convolution neural network is established.Firstly,the PC-Crash simulation software is used to build the models of cars and trucks,and their various collision conditions are simulated.The damage coefficients of the two vehicles after collision under the corresponding conditions are calculated,thus the data set of the simulation data needed by the deep learning neural network is obtained.Among them,the mass,relative speed,relative angle and relative distance of the two vehicles are the inputs of the neural network,the damage degree of the two vehicles after collision is the output of the neural network,and the damage degree is based on the deformation degree of the two vehicles.(2)Establish a deep convolution neural network and train,validate and test the simulation data set.According to the trained neural network model,the collision damage degree of vehicles and members can be predicted in near real time in a short time.The essence of this process is to use deep learning algorithm to predict text.In this paper,we choose the transfer learning method and use the parameters of the neural network trained from the data set of Boston House Price Forecast as the parameters of the convolution neural network,and construct the structure of the deep convolution neural network.(3)Combined with monocular deep learning algorithm,the above algorithm is verified by experiments.Experiments show that the algorithm can accurately predict the degree of vehicle damage after future accidents according to the current speed,other vehicle speed and driving direction on the road.In this paper,a video is intercepted,and the prediction accuracy is about 80% based on the trained neural network model.In summary,this paper designs a data-driven predictive anti-collision early warning algorithm,which uses deep learning algorithm to predict the damage degree of two vehicles after collision.It can accurately predict the damage degree of the vehicle after the accident in advance,so as to remind the driver to take corresponding actions in advance before the accident occurs,so as to avoid or reduce the damage degree of both sides.? This has important theoretical and application value for reducing the occurrence of car accidents and improving the safety of automatic driving.