A Safety Evaluation of Lane Change Based on Relevance Vector Machine

The accuracy and reliability are the major problems that have existed in traditional of lane change warning algorithms.Therefore,a test vehicle is set by using millimeter-wave radar,AWS vision sensor,vehicle gyroscope and other devices aiming at solving the above-mentioned problems.A field tests is conducted on highways and 19 subjects are recruited.Nearly 1 000 lane change samples are extracted from recorded driving test data.Based on a three-stage lane change trajectory model,the statistical value of yaw rate is regarded as the basis to determine radius of curvature in each stage.The upper limit for acceptable safety threshold is determined according to the yaw rate quantile when α=0.05.A series of control points are formed by analyzing lane change processes for which lasted more than 12s and the maximum value of lane change duration is selected as the benchmark.B-spline curve planning is adopted to determine the lower limit of the acceptable safety threshold.Parameters in lane change are estimated by Relevance Vector Machine(RVM),and 7th polynomial model is used to fit lane change trajectory.The area enclosed by the fitted trajectories and the upper(or lower)limit of the acceptable safety threshold is used as the warning parameters.The ratio is used as a warning parameter to evaluate safety of lane change.Actual data is used to verify this algorithm.The evaluation results of this algorithm can reflect actual safety level of the lane change,and have strong correlation with operating behaviors of drivers.