Study On Vehicle Collision Avoidance Warning Model Based On Lane-Changing Intention Recognition Of The Front Vehicle And Vehicle-to-Vehicle Communication

The cutting-in behavior of the adjacent vehicle in front will have a great impact on the driving safety of the backward vehicle,which is likely to cause accidents including side collisions and rear-end collisions.While the vehicles in front often suddenly cut in,the turn-on rate of turn signals in China is generally low,and the existing advanced driving assistant system(ADAS)has some problems in this situation,such as low warning accuracy and untimely warning.Therefore,this thesis relies on the support of National Key R&D Program of China: Development and Application of a New Type of Multi-functional Intelligent Vehicle Terminal(2018YFB1600701),and takes advantage of the driving simulation data to conduct an indepth study on the lane-changing behavior of the front vehicle.According to the lane-changing intention recognition of the front vehicle and vehicle-to-vehicle communication,the vehicle collision avoidance warning model is proposed,which will be helpful for the backward vehicle to acquire the timely and effective warning.Firstly,based on the driving simulator,a data acquisition test platform is built to collect the driver’s operating information and the vehicle’s dynamic parameters during the lane-changing process,which aims to establish a database on lane-changing driving intention.And using the characteristic driving data captured by the sliding time window method as input,the BiLSTM-based lane-changing intention recognition model is set up to recognize various driving intentions on different situations,including the left lane-changing,lane-keeping and right lanechanging.The experimental results about the training and testing of the model on the data set show that,the Bi-LSTM-based lane-changing intention recognition model in this paper has higher accuracy and could meet the real-time requirement in contrast to the BP and LSTM neural network models.Secondly,the data of lane-changing trajectory are extracted to analyze the influencing factors of lane-changing trajectory prediction to provide reference for the selection of characteristic parameters.Besides,the GRU neural network model is identified for use to predict the lanechanging trajectory in the light of some traits including the time-sequence of lane-changing trajectory and the strong feature extraction capability of deep learning.Using MAE and MSE evaluation indicators to assess the accuracy of trajectory prediction,the verification result finds that the error of the predicted trajectory in this prediction method is smaller than other neural network models.Thirdly,according to the analysis on the working conditions of the lane-changing and cutting-in of the adjacent vehicles in front,the overall framework of the collision avoidance warning model in view of vehicle-to-vehicle communication is put forward.After the backward vehicle obtains the intention of the front vehicle to change lanes through vehicle-to-vehicle communication,the lane-changing trajectory of the front vehicle will be predicted.Moreover,the bounding box theory and the separation axis theorem are utilized to detect whether the two vehicles will collide or not in accordance with the dynamic driving information of the backward vehicle,so as to propose the calculation method of time to collision of side(TTC-S)to make risk assessment.And after calculating and analyzing the time to avoidance(TTA),the grading warning strategy is made well by synthesizing the data,containing TTC-S,TTA and time of the driver’s reaction.Finally,the simulation scenario is designed for the working condition,on which the adjacent vehicle in the front is likely to change lanes to cut in.And a simulation experiment platform based upon the dual driving simulator is built to make the front vehicle and the backward vehicle be all operated by real drivers.Also,the proposed warning model and the traditional warning model of TTC have been simulated and verified respectively.The experimental results show that,the collision avoidance warning model set up in this paper could be accurately aware of the lane-changing intention of the front vehicle,and offer timely warning signals through trajectory prediction and collision detection to avoid collision.And another dimension to the results is that,this model with more timely warning has more successful in collision avoidance warning compared with the traditional one.