Pedestrian Detection Using Regression-Based Feature Selection and Disparity Map

Abstract

In this paper, the pedestrian detection using a regression-based feature selection and a disparity map method is proposed for improving the processing speed. Using many features helps to improve detection performance, but slows down processing. Therefore, it is important to select and use features efficiently. Our proposed method consists of three stages, such as a disparity map-based detection stage, a segmentation stage using a transformed disparity map, and a recognition stage with regression-based feature analysis. Through experiments with the ETH database, we show that the proposed method improves detection performance and especially processing speed.