Object Motion Analysis and Prediction in Stereo Image Sequences

Future driver assistance systems will have to cope with complex traffic situations, especially at intersections. To detect potentially hazardous situations as early as possible, it is therefore desirable to know the position and motion of oncoming vehicles for several seconds in advance. For this purpose, we propose a combined approach that tracks the vehicle position and orientation over time based on a box model, where the vehicle motion state is predicted several seconds ahead based on simultaneous tracking of multiple hypotheses with a particle filter framework. The scene is observed by a stereo camera mounted on the ego-vehicle. Compared to a traditional constant acceleration and curve radius prediction model, we show that the accuracy of the proposed particle filter approach is superior during turning manoeuvres displaying complex motion patterns.