Research On Human Action Recognition Of Somatosensory Interactive System

With the development of artificial intelligence, human-computer interaction becoming more user-friendly, traditional push-button mode or touch interaction way can not fully meet users’ needs any more, users are eager to have a non-contact way to offset the defects from the experience of current interactive products. As a natural human behavior, human body movement has become a popular aspect for scholars to study somatosensory interaction. Somatosensory interactive allows users to operate more convenient, more immersive, experience better. Somatosensory interaction is the development trend of interactive mode in the future.Researching on somatosensory technology of somatosensory interactive system is particularly important.So the main topic of the paper is researching on human action recognition.This paper summarizes the shortcomings of current development of human motion recognition based on kinect depth image and the problems includes action recognition instability, lack ing of dynamic training mechanism, recognition rate not high, the approach of rigging model making user fatigue. To improve these problems,the writer designed a human action recognition program by researching a lot of somatosensory interaction and motion recognition literature.And the program is applied in somatosensory interactive system that can improve well above problems. The program flow is as follows: 1. Obtain human skeleton image and use the proposed endpoint detection algorithm to extract motion data which constituted by human joints x, y, z coordinates. 2. Preprocess human joints x, y and z coordinates data and remove noise by using the moving weighted average method. 3. Regard the human joints x, y and z coordinates as the initial feature vector and Correct the feature model by using neural networks. 4. Models position track of human joints by using gauss and train the gaussian mixture hidden markov models through training samples.Finally, it proves the accuracy, timeliness and good interactive experience of this interactive somatosensory control system through lots of experiments and data analyses.