A two-layer switching based trajectory prediction method

Safety-critical situations in road traffic often result from incorrect estimation of the future behavior other users. Therefore, many Advanced Driver Assistance Systems (ADAS) need prediction models to ensure safety. Physical offer advantage general use and work quite well for short horizons, while longer periods time, maneuver-based better performance which, however, strongly depends on data used train them. An additional challenge is fact that surrounding can change its path, i.e. safety not only one maneuver should be considered but regular updates are required. Against this background, we propose a method uses three physics-based predictions – corresponding different assumptions combined with possible trajectories derived environmental knowledge. Continuous monitoring select most likely models. This choice then influences environment-based output both fused afterwards. The resulting Multiple Model Trajectory Prediction (MMTP) has been validated measured two scenarios city junction highway good without special measurements as commonly required prediction.