An adaptive approach to vehicle trajectory prediction using multimodel Kalman filter

With the aim to improve road safety services in critical situations, vehicle trajectory and future location prediction are important tasks. An infinite set of possible future trajectories can exit depending on the current state of vehicle motion. In this paper, we present a multimodel-based Extended Kalman Filter (EKF), which is able to predict a set of possible scenarios for vehicle future location. Five different EKF models are proposed in which the current state of a vehicle exists, particularly, a vehicle at intersection or on a curve path. EKF with Interacting Multiple Model framework is explored combinedly for mathematical model creation and probability calculation for that model to be selected for prediction. Three different parameters are considered to create a state vector matrix, which includes vehicle position, velocity, and distance of the vehicle from the intersection. Future location of a vehicle is then used by the software-defined networking controller to further enhance the safety and packet delivery services by the process of flow rule installation intelligently to that specific area only. This way of flow rule installation keeps the controller away from irrelevant areas to install rules, hence, reduces the network overhead exponentially. Proposed models are created and tested in MATLAB with real-time global positioning system logs from Jeju, South Korea.