Fast Pedestrian Detection Using a Night Vision System for Safety Driving
Mira Jeong, Jun-Yong Kwak, Jung Eun Son, Byoung Chul Ko, Jae-Yeal Nam · 2014
This paper proposes a rapid pedestrian-detection algorithm for thermal images by using energy symmetry and oriented center-symmetric local binary pattern (OCS-LBP) features based on luminance saliency. During preprocessing, energy symmetry based on luminance saliency is used as the filter to remove objects and to reduce the pedestrian classification time. The OCS-LBP feature is then extracted from a candidate window and subjected to a random forest classifier (RF) that separates candidate windows into pedestrian and non-pedestrian classes. The proposed algorithm has been successfully applied to various thermal images captured in a car, and its detection performance is better than that of other methods.