The gradient based feature, such as histograms of oriented gradients, focuses on the spatial distribution of edge orientations, but disregards the color information. Color-based features are very popular in image classification but rarely used in human detection. In this paper we propose a new human detection method by combining texture-based features with color information. Basically, local binary pattern (LBP) is used as a texture feature, and a new color feature, relational color similarity (RCS), is introduced to enrich the descriptor set. By combining RCS and LBP as the feature set, adopting linear support vector machine (SVM) as the classifier, carefully designed experiments demonstrate the superiority of RCS-LBP over other traditional features for human detection on INRIA human database.