3D Human Body Model Based On Monocular Image Motion Gesture Recognition

With the Developing of the virtual reality technology, more and more people will be virtual reality and computer vision research together to explore solutions to some of the field of computer vision problems. The model-based visual analysis of image motion long before people had begun to study, but mostly based on two-dimensional human body model, two-dimensional human body model because the lack of dimension, making the image analysis of human motion in time will resulting in loss of information, there are significant limitations.This study is intended to combine virtual human motion control technology,and the use of three-dimensional human body model to match images of the human body posture recognition. This study is divided into two major parts, one for three-dimensional human body model of real-time motion control, the other part of the image processing and matching. Gesture interaction of these two parts identification, three-dimensional human body model in real-time adjustment of the control part of the three-dimensional model to adjust the movements and appearance, the resulting projected image is passed to the image processing and matching module, the image processing and matching module will receive the projected image and first to identify the image pre-processing, using finite difference method to obtain the body target area, then the source image to use as a template matching and edge matching gesture area on the projected image to do the evaluation function evaluation, based on the evaluation results of three-dimensional human body model to estimate the next movement state set of parameters, the three-dimensional human body model by the feedback control part of the set of parameters to drive the three-dimensional human body model for the next attitude adjustment process until the entire cycle of attitude evaluation function to reach convergence, three-dimensional human body model of projection and source images of the human body posture similar information to complete the identification.we use different actions Weizmann database image to be identified, the results show that our method for a wide range of physical activity has a high recognition rate.