3D vision sensing for improved pedestrian safety

Abstract

Pedestrian-vehicle accidents account for the second largest source of automotive related fatality and injury worldwide. This paper presents a system which detects and tracks pedestrians in realtime for use with automotive Pedestrian Protection Systems (PPS) aimed at reducing such pedestrian-vehicle related injury. The system is based on a passive stereo vision configuration which segments a scene into 3D objects, classifies each object as pedestrian/non-pedestrian and finally tracks the pedestrian in 3D. Our system was implemented and tested on a Volvo test vehicle. Strong results for the system were obtained over a range of simple and complex environments, with average positive and false positive detection rates of 83.5% and 0.4% respectively.