Research and Design on Vehicle Pedestrian Detection in Natural Scene

Abstract

We propose a two-layer pedestrian feature extraction algorithm based on multi features fusion in integral channel, which is gained by intelligent driving system in natural environment. The algorithm extracts the gradient direction, gradient magnitude and LUV color channel of the image as upper layer feature, by using the fast operation speed of integral channel. Edgelet features that describe pedestrian local information are used as lower level features for validation. While ensuring real-time detection, we improved the low robustness of single layer features. Experimental results show the algorithm can reduce the false detection rate to a great extent under the premise of ensuring the detection speed in dealing with pedestrians in the background of natural and complicated vehicles.