A stereo cameras setup for pedestrian detection enhancement

Abstract

Pedestrian detection is an important and challenging area in computer vision with the potential to save lives. Although many works have already been developed to improve performance of pedestrian detection algorithms, none have yet been able to handle detection from afar due to pedestrians' small scales on images. This prevents these algorithms to be reliably applied to autonomous vehicles to avoid accidents on fast moving traffic. In this context, this paper proposes a cameras setup, which consists of two stereo cameras with different focal lengths and baselines, allowing to have higher pedestrian resolutions on images for a larger range of distances in front of the vehicle. Experimental results reveal a considerable enhancement on the detection performance, overcoming the difficulty caused by the reduced scale that pedestrians have on images.