Research On Human Action Recognition Method Based On Stereo Vision

The interaction between human and computer becomes more and more closely during thetime of high-speed development. It is mainly reflected in scientific research, sports, medical,transportation and other fields. People get information, perform scientific research andexperience digital entertainment by operating the computer. At the same time, a new researchfield is put forward: human-computer interaction. And human action recognition technology isthe foundation of human-computer interaction. The technology has been mature because of thesocial needs of all walks of life and also widely used in intelligent monitoring, intelligentanalysis in areas such as health care, sports. It has a very broad application prospects andconsiderable economic value. The video monitoring system has replaced the expensive humanmonitoring and realized the monitoring in dangerous environment or that human cannot beachieved in reality. For intelligent medical field, it can avoid the waste of a lot of manpower ormaterial resources, at the same time, reasonable and standardized rehabilitation training also canmake the patients get better faster. Based on the analysis and summary on the basis of existingtheory and research, this paper proposed human recognition method based on the LCS algorithmand body posture recognition based on SVM algorithm.This paper adopted Microsoft’s new product—Kinect body sensor. It is adopted by manyresearchers to study the field of action recognition and face recognition because of its low priceand convenience. This paper mainly studied the following several aspects based on thethree-dimensional coordinate values of human bone joints.(1) The definition of the human action. For users’ posture of a frame or a continuous periodof time sequence, how to get the data of human skeleton and extract the feature that representsthe characteristic of continuous movements is one of the main research contents in this paper.(2) In the process of human body posture recognition, the use of the SVM classifier andhow to design an efficient classifier according to different posture is also one of the mainresearch content of this paper.(3) The process of human action recognition is to get the real-time action features of usersand compare with the saved feature sequences of all defined actions by LCS algorithm, and thenfind the longest LCS. At last, we can identify the action of human body. At the same time, howto improve the LCS algorithm and the operation efficiency are the main issues of this paper. Through the result and analysis, the algorithm has high accuracy and strong robustness.Human action can be accurately identified.