PEDESTRIAN DETECTION USINGSTEREO VISION AND TRACKING

摘要

This paper presents a method of pedestrian detection for automotive applications. A visionrnbased pedestrian detection presents a difficulty due to the diversity of pedestrian appearance.rnThe proposed method overcomes this difficulty, through the following two contributions.rnFirstly, employing four directional features with the classifier increases the robustness againstrnsmall affine deformation of objects. Secondly, by merging classification and tracking,rnrobustness with respect to temporal changes in appearance is improved when consideringrntemporal continuity of a classification score. Experiments, performed on approximately 16rnminutes of video sequences, confirmed that the method can detect pedestrians with low falserndetection.