Motion Recognition System Based On Multi-modal Data Features

Motion recognition has been an important part of computer vision since its appearance and the achievement of it could be applied to various areas such as security system, smart home, somatosensory interaction and so on. Multifarious devices such as kinect, psmove, wii for the research of motion recognition emerge in endlessly with the development of technology. Meanwhile, science researchers have tried their best to develop all kinds of algorithms to tackle with the data collected by somatosensory devices and then perform motion recognition on these datasets.In this paper we have done research on the topic "motion recognition based on both skeleton and depth data" and then come up a novel method for motion recognition. After that we realize the data collection system and motion recognition system. We basically complete three primary tasks in this paper:first, we come up with a new kind of classifier called body part feature fusion classifier. Second, we introduce the concept of motion recognition based on multi modal data. When our body appear behind some obstacles such as a chair or some other people the data captured by somatosensory devices will not always be correct, which affects the result of motion recognition. As a result, we use the depth data as a complement in case of such circumstance because depth data remains stable. third, we realize the data collection system and motion recognition system. we first gather skeleton data and depth data from kinect and then process these data separately, finally we use support vector machines to make a classification when a new part of motion segment which is unknown came into our system.The new feature that we have came up with is a complement of the existing features, which utilize the body data more adequately and hence it increase the accuracy of recognition. The algorithm that we have came up with combines the advantage of recognition with skeleton data and the advantage of depth data, which will also increase the recognition accuracy as a result. In a word, all the efforts we have made is profound and will be utilized by scientists for further study in the near future.