Abstract: On the demand of hand interaction in the head-mounted display environments of virtual reality systems,a novel method of hand pose estimation is proposed in this paper.The proposed method converts hand tracking problem into a hand pose indexing problem through locality sensitive hashing(LSH).In this paper,the histogram of oriented gradients(HOG) is served as the indexing feature,which could be indexed by an improved multi-probe scheme with high efficiency.As the results of indexing are weighted by temporal consistency,the hand pose configuration including finger joints and wrist without markers could be estimated.Our experiments of applying the proposed method on a virtual cockpit system verify the its practicability and demonstrate its robustness on the hand self-occlusion problem.