SVM Based Trajectory Predictions of Lane Changing Vehicles

Abstract

In dense traffic, a lane change maneuver is potentially dangerous and a driver's erroneous estimation of inter vehicle gaps may cause an accident. This signifies the necessity of forewarning the driver of the feasibility of lane change. This requires an early prediction of the trajectory of vehicles. A lane change has three distinct phases, planning phase, lane change phase and the adjustment phase. If the lane change intent of a driver is discovered in the first phase, he can be suitably advised. Else, the prediction of an impending collision through long range trajectory prediction would require a reactive approach. In this paper, we present support vector machine (SVM) based prediction of the trajectory of lane changing vehicle. SVM is used for both short range and long range trajectory prediction of the LC vehicle. A vehicle trajectory is modeled as a discrete time series. Sufficient number of values is taken to include planning, lane change and the adjustment phases of the time series in the prediction. The SVM based prediction is performed using actual Next Generation Simulation (NGSIM) field data. SVMs outperformed earlier algorithms and proved especially effective in early detection of driver lane changes on the road.