Vehicle Swarm Collision Avoidance Warning Algorithm Based On Driving Intention Sharing In Vehicle To Vehicle Environment

In recent years,the overall situation of transportation safety in China has been improved,but due to the frequent traffic accidents caused by drivers’ untimely response to emergencies and inappropriate emergency handling.With the issuance of a document aims to set up transport networks with wider coverage and higher speed in 2019,higher requirements have been put forward for "improving the level of intrinsic safety".With the rapid development and application of Intelligent Vehicle Infrastructure Cooperative Systems in active traffic safety and vehicle collision avoidance,it provides a new research idea and direction to achieve this goal.Vehicle to Vehicle(V2V)communication has the technical characteristics of low delay,fast access,high transmission,high stability and so on,which increases the vehicle perception range of surrounding information,and information sharing can be realized among multiple vehicles.The vehicle can obtain more timely,reliable and comprehensive motion state and position information of other surrounding traffic participants.Compared with the traditional collision avoidance early warning algorithm,it can enhance the early warning effect in non-line-of-sight.This paper mainly takes the vehicles on the real road under the V2 V environment as the research object,focusing on three aspects: driving intention prediction,vehicle trajectory prediction and collision avoidance early warning algorithm design.Firstly,a field vehicle experiment in V2 V environment is designed and completed to collect the vehicle geographic information,vehicle motion state information,and driving environment information.Cubic spline interpolation is used to repair the data,Ttest is used to correct abnormal data values,Kalman filtering method is used to filter noise and correct jumps.Vehicle "lane change" and "car following" data are extracted to provide data support for driving intention prediction and trajectory prediction.Secondly,a driving intention prediction model based on Hidden Markov Model(HMM)is established to solve the prediction intention with the observable vehicle motion state.A lane change MGHMM driving intention prediction model is proposed,which takes vehicle azimuth,angular velocity,lateral speed,lateral acceleration and lateral coordinates as observation variables to predict lane keeping and left or right lane changing intentions.At the same time,a car following MGHMM driving intention prediction model is proposed,which takes vehicle longitudinal speed,longitudinal acceleration,and the headway relative to the vehicle in front as the observation variables to predict the intention of constant speed,acceleration,deceleration and emergency braking intentions.The results show that the prediction accuracy of the lateral lane change intention is over94%,and that of longitudinal car following driving intention is over 87%.Then,taking the driving intention prediction result and the vehicle history trajectory as the model input,the MGHMM-LSTM vehicle trajectory prediction model considering the driving intention is established.Through the prediction of the vehicle lateral and longitudinal acceleration,the predicted vehicle lateral and longitudinal trajectories are calculated.The results show that the mean absolute error of longitudinal coordinates of the car following trajectory prediction model is 0.028,and the root mean squared error is0.047.The mean absolute error of longitudinal coordinates of the lane-changing trajectory prediction model is 0.081,and the root mean squared error is 0.189.The mean absolute error of lateral coordinate prediction is 0.053,and the root mean squared error is 0.108.Compared with the traditional trajectory prediction model,good results have been obtained in vehicle trajectory prediction.Finally,based on the Separating Axis Theorem and Intention Conflict Theory,a Driving Intention based Vehicle Swarm Collision Avoidance(DI-CA-SWARM)warning algorithm is proposed.The proposed DI-CA-SWARM early warning algorithm is verified by the real vehicle cluster experiment.The early warning algorithms,whether considering the driving intention or not,are compared and analyzed from the three aspects of correct rate,false alarm rate and missed alarm rate.In this paper,the results show that the proposed algorithm can accurately predict the collision risk and warn the vehicle in advance.Compared with the early warning algorithm without considering driving intention,the average early warning rate increases from 86.29% to 91.46%,the average false alarm rate decreases from 8.46% to 6.03%,and the average missed alarm rate decreases from 5.26% to 2.51%.