Human Motion Capture Based On Depth And 3D Model

The thesis has carried on the analysis of motion capture algorithm and technology based on existing researches. In this thesis, in order to overcome some disadvantages of existing methods, the motion capture method based on depth images and 3D models is proposed. First, it gathered the depth images by Kinect. Second, it removed the background of depth image and recovered the 3D models from depth image so as to establish the database of three-dimensional model. Third, it extracted skeleton of body using depth image so as to create database of skeleton. Four, it input a set of depth image sequence of motion so that it could match with the 3D model of database. After calculating the distance between two motion sequences, the corresponding skeleton would be output as the result of motion capture. The specific work of this thesis is showed as the following aspects:(1) The thesis proposes the research direction after summarizing the background and status of motion capture.(2) The thesis proposes to obtain the depth image using kinect after researching the mainstream methods. At the same time, iterative threshold is proposed to remove the background of depth image.(3) The thesis proposes to reconstruct 3D model based on depth image. Using the three dimensional information of depth image can recover 3D models conveniently. At the same time it can get the point cloud data of 3D model.(4) The thesis proposes to extract skeleton using depth image. Firstly, it finds the points in skeleton, and then it connects the points with simple lines. This algorithm extracts the skeleton successfully.(5) The thesis determines using the recognition method to carry on the simulation and analysis of motion capture. It establishes two corresponding databases of 3D model and skeleton. Otherwise, the thesis determines using the hierarchical clustering algorithm to cluster the point cloud data of 3D model. At the same time, the thesis suggests measuring the distance of motion sequences with DTW, and the experiments have confirmed the validity and feasibility of the algorithm.