A Controlled Interactive Multiple Model Filter for Combined Pedestrian Intention Recognition and Path Prediction

We present a novel approach combining pedestrian intention recognition and path prediction for advanced video-based driver assistance systems. The core algorithm uses an Interacting Multiple Model Filter in combination with a Latent-dynamic Conditional Random Field model. The model integrates pedestrian dynamics and situational awareness using observations from a stereo-video system for pedestrian detection and human head pose estimation. Evaluation of our method is performed on a public available dataset addressing scenarios of lateral approaching pedestrians that might cross the road, turn into the road or stop at the curbside. During experiments, we demonstrate that the proposed approach leads to better path prediction performance in terms of a smaller lateral position error compared to state-of-the-art pedestrian intention recognition and path prediction approaches. The computational costs of our approach is comparatively low and therefore can be ported easily onto a real-time system.