Edge based segmentation for pedestrian detection using NIR camera

In this paper we present a fast and robust segmentation scheme for a nighttime pedestrian detection system. The system uses an IR source for illumination and (Near Infrared) NIR camera to capture images of a scene at night. A new vertical edge detection method is used to identify edges in images that belong to pedestrians. These edges are further combined to form potential pedestrian image blocks called as `candidate blocks'. Tight bound rectangular blocks are obtained from the candidate blocks using intensity profiling. The candidate blocks also undergo a process of elimination based on certain criteria, in order to reduce false positives. Further, two different types of schemes are experimented for tracking of the pedestrian blocks. One is based on template matching and the other is based on image segmentation. Performance characterization of the algorithm has been carried to evaluate its robustness against contrast variations. The experimental results show that the algorithm is robust and fast, for use in real-time applications. Invention disclosure has been filed for the method described here [9].