Inspired by Weber's local descriptor and LBP feature ， this paper proposes a SLBH (Saliency Local Binary Haar) feature in view of the weaknesses that Haar features have high dimension and redundancy. SLBH helps obtain good detection performance when using the overall characteristics of pedestrian ， but the detection performance declines rapidly in occlusion scenes. In order to improve the robustness of overall characteristics to partial occlusion ， a two-stage pedestrian detection algorithm combining multi-component validation and SLBH is proposed ， which takes the advantage of overall feature and local feature simultaneously ， and improves the robustness of the algorithm to partial occlusion. Experimental results on INRIA pedestrian detection dataset show that the proposed algorithm is of strong robustness to noise and partial occlusion.