Multiple Model based Lane Change Prediction for Road Vehicles with Low Cost GPS/IMU

This paper proposes an interactive multiple model (IMM) based method for predicting lane changes in highways, as a part of a scene interpreter system. The sensor unit consists of a set of low cost GPS/IMU (Global Positioning System/Inertial Measurement Unit) sensors and an odometry captor for collecting velocity measurements. Extended Kalman filters running in parallel and integrated by an IMM based algorithm provide positioning and maneuver predictions to the user. Two different maneuver states, change lane (CL) and keep lane (KL), are defined by two models describing different dynamics. Real trials in highway scenarios show the capability of the system to predict lane changes in straight trajectories or those with low curvature. Further investigations will be dedicated to apply this method to curved highway stretches. The results presented in the paper show the suitability of this option in the cases analyzed, with very short latency times. The integration of GPS and Micro-Electro-Mechanical (MEM) inertial sensors in the onboard unit has been proved to be capable of providing uninterrupted positioning and an efficient lane change predictor at low cost.