Research On The Method Of Time To Line Crossing Prediction Based On Driving Intention Identification

With the rapid development of China's road transport industry and vehicle ownership,frequent occurrence of traffic accidents lead to property losses and casualties,which seriously affect people's production and life.Advanced driver assistance system is used to monitor the inside and outside environment around vehicle with the help of a variety of sensors installed on it.In an emergency,the algorithm can alert the driver and even actively interfere vehicles to avoid traffic accidents based on preset algorithm strategy.Current lane departure warning system of ADAS only adopt time to collision(TTC)as a single indicator to assess the risk of lane change,resulting in a high false alarm rate.Time to line crossing(TLC)is proposed to use to detect driver's lane change behavior in this paper and the system only gives a warning when TTC and TLC meet thresholds set by the lane change warning algorithm simultaneously.The departure warning decision based on TLC is highly dependent on the external environment perception and vehicle steering signal.The activation of the warning system is contrary to driver's intention causes dangerous accidents also.Therefore,it is essential to perceive driver's operating behavior,identify its driving intention accurately,and carry out further research on the departure warning system based on TLC and TTC for the system.Firstly,This paper analyzes and summarizes research findings on identification of driving intentions and the lane departure warning system based on time to line crossing.It makes a comprehensive analysis of“human-vehicle-road”system parameters in vehicle driving based on the division of driver's visual field and characteristic parameters which can represent different intentions.It is imported into Hidden Markov Model and Support Vector Machine to identify driving intention.Finally,the lane-change left department is chosen to analyze.The influence of parameters such as lateral acceleration,speed and yaw rate are fully considered,and the geometric kinematic lane changeover model of the vehicle is established to predict the time to line crossing.The main contents of this paper are as follows:(1)Characteristic parameters selection of human-vehicle-road system.Firstly,the K-means dynamic clustering method is used to divide the interest area of driver's line of sight,and the abnormal sightline is removed by using the modified Pauta rule to clarify driver's field of vision focus area:the dead ahead,left and right front of the lane,left and right rearview mirror.Statistical analysis verifies the differences of driver's visual parameters,using R-type index clustering to determine driver's visual characterization parameters:v=[GA_f GA_t GL_a GL_p H_a].Analyzing parameters of“Vehicle-Road”system,and determining longitudinal acceleration A_x,steering wheel angle S_a and distance between the vehicle and the lane center-line D_y as characterization parameters.(2)Driving intention identification based on HMM and SVM.Experiments were conducted in the simulator-driving system.1150 driving samples of 12 subjects were recorded,after that the collection and selection of parameter information for various intentions.Kalman filter is used to preprocess the longitudinal acceleration and steering wheel angle.Based on principal component analysis method,dimensionality reduction of characteristic parameters sequence was performed to determine 4principal component sequences that characterize driving intentions.The corresponding principal component sequences are imported into HMM model to recognize the intention of following and right lane-change.The main component sequence corresponding to the ambiguous left lane-change and overtaking intention with low recognition rate in HMM is achieved that as a candidate set to be identified,the second-level SVM is introduced to make a final decision on the identification intention.The results show that recognition accuracy rate based on this algorithm is95.84%,which is obviously higher than HMM or SVM single model,its recognition time is 0.017s meeting with driver's reaction time requirement for sudden events.(3)Time to line crossing model establishment based on vehicle movement geometry analysis.Combining the motion characteristics of the vehicle with the trajectory characteristics of the road environment and traversing track,In order to estimate the time to line crossing(TLC),several straight(curved)line sections along the straight(curved)track are built to estimate TLC.Based on Matlab software,it is carried out that simulation analysis of parameters affecting the crossing time TLC,which plays a guiding role in the driver's actual driving operation process.The validity and rationality of the model are demonstrated by cross validation of driving simulator and real vehicle test.The result show that the time to line crossing prediction model established in this paper is more reasonable.