Skeleton Tracking And Motion Recognition Based On Kinect

With the continuous innovation of science and technology, people are getting higher and higher requirements for intelligent technology, The motion recognition technology of the human-computer interaction has become an important research direction, which makes our devices more intelligent, and Microsoft launched the Kinect device is to accelerate the development of interactive systems. It can capture the human depth images, color images and skeleton images, we need to collect these data for processing and application, it will be practical to be used in people’s daily and people’s action. Nowadays, many of the research based on motion recognition is based on video or monocular camera, the real-time performance of the system is very poor, when the application in human-computer interaction is even inflexible, and with the advent of the Kinect, we wouldn’t have to worry about this problem, we can through the Kinect in real-time acquisition to recognize motion, through the computer identification to realize the control, so the skeleton tracking and motion recognition technology is the key research direction.This paper briefly introduces the Kinect hardware composition and existing action recognition related theory knowledge, and realizes real-time skeleton tracking and the motion recognize.This paper has realized the PPT turning through motion recognize, then the recognition effectiveness is verified, and we also study it through different way for different complexity of the action. The main research work are as follows:(1) Firstly, describe the background and significance of the Kinect motion recognition systems, through the domestic and foreign research literature and research status of laboratory, we understand the methods of motion recognition based on different ways and analyze the advantages and disadvantages of the methods, and identify the overall framework of the motion recognize system.(2) The human skeleton tracking is realized through the Kinect, Kinect hardware structure and working principle are introduced in this paper, first of all to describe a extraction method of the depth of the target image, and the preprocessing of the depth image, to realize the accuracy of the depth image extraction, and then introduce the Kinect skeletal points extraction principle, it is accomplished real-time skeleton tracking through the experiment.(3) Research on the motion recognition through two different ways, one of this is defined by physical method to achieve real-time control of the simple action, through the realization of Kinect on human skeleton tracking and recognition of the action, it completes the interaction system to control the office software, through a better way high to improve the recognition rate of system; another method is to recognize through the template matching, firstly we should create motion database and database index by Kinect, extract the quaternion of the skeleton and establish its feature database and get the corresponding feature database, and then by the Euclidean distance of the matching algorithm to match the recognition, before the Euclidean distance, we need to reduce the dimension of the algorithm for its data integration to get a similar action.(4) This part tests the system through a lot of experiments, compares the recognition effect by the different people and the different parameters, and verifies the proposed method.Through a large number of experimental results and data analysis show that when we choose the appropriate parameters of the system identification effect will be improved greatly, and different actions that be defined have their advantages and disadvantages.