The combination of CSLBP and LBP feature for pedestrian detection

Local Binary Pattern (LBP) feature has attracted increasing interest in pedestrian detection tasks, but it requires high dimensionality to gain promising performance. To address this problem, we make three main contributions. First, Center-Symmetric Local Binary Patterns (CSLBP) feature is introduced to pedestrian detection. It´s the first time to extract CSLBP operator in an approach similar to basic LBP operator. Second, we propose the operator combining CSLBP_1,8 pattern and uniform LBP_1,8 pattern, named combined-LBP descriptor for convenience. Third, the combined-LBP operator is applied to various color spaces. Experiments on INRIA pedestrian database show that combined-LBP operator based pedestrian detector achieve state of art performance with miss rate of 6.04% at FPPW=10^-4 in gray-scale image, and the proposed operator in oRGB color space obtain superior result with miss rate of 2.75% at FPPW=10^-4.