Research On Side Collision Risk Warning Method In The Vehicle To Vehicle Communication Environment

Advanced Driver Assistant Systems(ADAS)has been shown to significantly reduce traffic accident caused by human significantly.However,due to the complex traffic environment,warning algorithms that rely solely on self-vehicle sensing information have false negatives and false positives problems.In the traffic accidents,side collision caused by sudden cut-in of other vehicles accounted for a certain proportion,while the existing ADAS has a low accuracy rate of warning in this scenario.In order to improve the effectiveness of ADAS in the cut-in scenario,this paper establishes a warning model of the side collision risk under the Vehicle-to-Vehicle(V2V)communication environment to improve the warning timeliness and accuracy.The research results in this paper can provide new modeling idea and theoretical basis for further optimization of ADAS in above scenarios.Firstly,build a real vehicle experimental platform,use the real vehicle to install sensors to collect vehicle kinematics data and driving environment data to simulate the V2 V communication environment.Recruit male skilled drivers to conduct experiments on urban expressways.After pre-processing,calibrating,and analyzing the data,210 lane-change events were screened as a library of lane change samples.Subsequently,based on the experiments data,a lane-change behavior prediction model and a vehicle trajectory prediction model were established.Based on Support Vector Machine-Recursive Feature Elimination(SVM-RFE)algorithm,the driver's lane change intention recognition model was established.The recursive feature elimination method was used to select features.The grid optimization method was used to find the optimal parameters.The model performance was evaluated on the independent test set by Area Under ROC Curve(AUC).The three conditions of recognition result of lane change intention,lane change feasibility condition and vehicle lateral offset change rate used logical “&” to establish a lane change behavior prediction model.Based on Long Short-Term Memory(LSTM),the vehicle trajectory prediction model was established,and the trajectory predictions of lane keeping and lane change were modeled separately.The vehicle trajectory prediction model under lane keeping conditions calculates the longitudinal displacement of the vehicle by the predicted longitudinal acceleration,and the vehicle trajectory is obtained by the longitudinal displacement.The vehicle trajectory prediction model under the lane change condition was divided into a longitudinal displacement prediction model and a lateral position prediction model,in this case the predicted trajectory of the vehicle is comprehensively obtained from the predicted longitudinal displacement and the predicted lateral distance.Then,based on the lane change behavior prediction model,the vehicle trajectory prediction model and the oriented bounding box detection algorithm,the side collision risk warning model in V2 V communication environment was established.Based on the lane change behavior prediction model,the lane change behavior of the target vehicle was judged per 0.5 seconds,and the judgment result was transmitted through V2 V communication.If the target vehicle was about to adopt the lane change behavior,the LSTM trajectory prediction model was used to predict the trajectory of the host vehicle and the target vehicle.Use the oriented bounding box detection algorithm to detect whether there is a collision risk between the host vehicle and the target vehicle at each predicted point.Finally,based on the real vehicle cockpit driving simulation platform,the cut-in scenario in the V2 V communication environment experiment was designed and implemented.The positive and negative samples were extracted by the expert evaluation method.The proposed collision risk warning model and the traditional collision warning model were respectively run on the continuous data of the samples.Finally,it is proved that the side collision risk warning model proposed by this paper is better than the traditional one,by the comparison and verification of the warning confusion matrix and the early warning time.