Skip to Main Content
This paper introduces a vision-based motion capture system. Motion capturing technology consists of two categories: model-based tracking and example-based indexing. The motion capturing systems face two challenges: parameter estimation in high-dimensional space and self-occlusion. Our algorithm extends the locality sensitive hashing (LSH) method to find the approximate examples and then estimates the pose parameters in high search space. The contributions of this method are proposing the modified LSH function, applying Hough voting to estimate the pose parameters, and adding the temporal/prediction constraints to increase the prediction accuracy.