Research On Vision Based Somatosensory Interaction Technology

With the development of intelligent life, all kinds of somatosensory interactive technologies are constantly emerging. Somatosensory interaction based on natural gestures conforms to the human habit of daily communication and has wide application prospect in Intelligent Appliances, Virtual Reality, etc. At present, however, most of the gesture recognition technologies are difficult to be applied for various reasons. For instance, the complexity of algorithm will lead to bad real time performance as well as the system is subject to background environment, user dress and postures.In allusion to the above problems, this paper presents a new real-time gesture recognition technology. It can be applied to different complex scenes, no matter whether there is skin color in the background, or user’s arm and face in the scenes. What’s more, the users even do not need to wear any props.First, we set up a new HLS-CbCr mixed color space to detect skin color, and use Gaussian mixture background modeling algorithm to detect foreground. With the help of skin detection results, we set the learning rate of model by introducing the concept of time and space information to improve the rate of algorithm. As for face and other interference problems, we propose a pre-detection algorithm, and then use a complementary mechanism to group various detection results to segment gestures.Secondly, in feature selection respect, we propose the use of combination of gestures image features, as well as choosing features of low computation complexity and high efficiency, including hu moments, defects and scale features.Finally, we use the Support Vector Machine (SVM) to recognize gestures, and build effective defect filters to point out the location of fingertips, which can improve the algorithm speed. Moreover, we combine these two results to further improve the recognition rate.According to experiments, the proposed algorithm has got stable results under different complex scenes. And the gesture recognition rate can reach97.02%. In the end, we design a virtual interactive system with gesture control, which can be applied to smart TVs, air conditioners, mobile phones and other electronic products, and can be extend to more application areas.