Sparse Representation-based Human Motion Capture Data Analysis Methods

In recent years, with the rapid development of CG industry, motion capture technology has become more and more mature. And it is more and more used in film making, games production, sports training, medical simulation, etc. However, due to the disadvantages of motion capture techniques, such as capture device tend to be expensive, the operation of the equipment is complex, and needs strict limits on scene, it limit the widespread use of the technology. How to reuse existing motion capture data to create more vivid and realistic 3D body animation, it is a hot issue in the field of this study at present.With the sparse representation theory being more perfected, the sparse representation of the signal gains concern from more and more scholars. It has been wildly used in various fields, and provides a good solution to various problems.Analysis and processing human motion capture data by using sparse representation theory to obtain the better effect, which is the goal of this thesis. The work of this thesis mainly includes the following aspects:(1) The segmentation algorithm of human motion capture data based on sparse representation. It is proposed under assumption of local linear relationship in human motion. The algorithm firstly extracted a subsequence as a dictionary from the front of the motion sequence for segmentation. The subsequence contains at least a movement cycle of a kind motion type. Then, it solves the sparse representation and reconstruction error of each frame in the following sequence on the dictionary. It segments the motion by detecting transition point in the error sequence. At last, the result of experiments shows that this method can obtain a better effect for segmentation.(2) The motion retrieval algorithm based on sparse representation. For a query motion sequence submitted by the user, the algorithm extracted the dictionary from it. The sequences which are similar to the user's input in the database can obtain smaller reconstruction error after they are sparsely represented on the dictionary. In contrast, the dissimilar sequences will get greater reconstruction error. Based on this assumption, this thesis proposes a content-based motion retrieval algorithm by using sparse representation theory. Through the experiment with the existing methods, it is demonstrated that this algorithm can get a higher retrieval precision.(3) Recovery of missing data base on sparse representation. Firstly, an algorithm of predicting missing markers in human motion capture using L1-sparse representation is studied. By using the framework of sparse representation theory and the continuity of human motion, the algorithm divided the missing marker posture into two parts:the complete part and the incomplete part. It solves the sparse representation of the complete part on the dictionary formed with the training set. And then, it reconstructs the incomplete part using the dictionary and the representation coefficient. According to the deficiency of the original algorithm, the linear quaternion is adopted as the feather, and it only select important joint in human body skeleton to solve sparse representation. In addition, in order to enhance the adaptability of the original algorithm, this paper present an improvement on the presentation coefficient weighted update algorithm. The experimental results show that the improved method not only has lower reconstruction error, but also can reduce the recovering time and improve the efficiency of the algorithm.