Dense stereo-based critical area detection for active pedestrian protection system

Abstract

An active pedestrian protection system is required to operate in real-time and produce an extremely low false alarm. To achieve these requirements, Broggi et al. detected critical areas using a laser scanner and found pedestrians only in these areas via a camera. However, using heterogeneous sensors requires a complicated system architecture and laser scanners have drawbacks in terms of installability and durability. Therefore, proposed is a critical area detection method using a stereo camera. This configuration can overcome the aforementioned drawbacks and detect both critical areas and pedestrians using a single sensor module. The proposed method detects critical areas by extracting local extrema of the depth change profile along a free space boundary. Experiments show its feasibility quantitatively.