Online inference of lane changing events for connected and automated vehicle applications with analytical logistic diffusion stochastic differential equation

•The stochastic lane-changing trajectory is modeled with the logistic diffusion method, which is represented by a stochastic differential equation. The analytical solution is derived.•The analytical distribution of the lateral movement is solved by the Fokker Planck equation.•A procedure of inferring lane changing initiation moment (LCIM), cross-lane-mark moment (CLMM), and lane changing duration (LCD) is established.•The proposed method is tested against real-world data. The results are encouraging. Inferences of the lane changing behaviors (including lane changing initiation moment LCIM, cross-lane-mark moment CLMM, and lane changing duration LCD) of the surrounding vehicles and taking actions to avoid collision are essential for the safe operation of connected and autonomous vehicles (CAVs). The majority of current models rely on data-driven methods that need to be trained before deployment. Besides, an analytical and stochastic formulation of the lateral model which can generate CLMM and LCD distribution and adjust the parameters online is still lacking. To the best of our knowledge, currently, no method can provide an analytically stochastic lateral movement that explicitly considers the system noise and simultaneously outputs the information of lane changing initiation moment, cross-lane-mark moment, and lane changing duration in a real-time manner. To fill such a gap, a stochastic lateral trajectory framework with parsimonious parameters is established and an online simultaneous inference of LCIM, CLMM, and LCD is developed. The proposed method is tested with the highD and NGSIM datasets. The results show that 1) computational efficiency-wise, the algorithm takes milliseconds to run, which suggests promising prospects for field deployment; 2) the false negative and false positive errors of the LCIM inference are as low as 1.5%, indicating that the method is robust to the stochastic noise; 3) LCIM inference is insensitive to the lane width, which means the model is error tolerant when such information is lacking; and 4) the error of CLMM inference is within 2 sec, while most of the error of LCD locates within 4 sec, suggesting satisfactory performance.