Occlusion-robust model learning for human pose estimation

Abstract

In this paper we examine the efficacy of self-occlusion-aware appearance learning for the part based model. Appearance modeling with less accurate appearance data is problematic because it adversely affects entire learning process. We evaluate the effectiveness of mitigating the influence of self-occluded body parts to be modeled for better appearance modeling process. To meet this end, We introduce an effective method for scoring degree of self-occlusion and we employ an approach learning a sample proportionally weighted to the score. We present our approach improves the performance of human pose estimation.