Optimized Algorithm Of Human 3D Motion Recognition Based On HOG And LBP

Abstract

Recognizing the accurate position of each joint point of the human body helps the computer to understand the rich human body motion information. Recognizing the precise position of each joint point of the human body in 3D space helps the computer understand the rich human motion information. In order to avoid the defects of binocular vision combined with neural network method, we have designed an optimization algorithm for human body 3D pose recognition technology under binocular vision. The algorithm uses HOG and LBP feature operators to perform feature matching on the corresponding joint point areas of the left and right images, and seeks the optimal pixel position of the corresponding joints points. And the search window can adjust itself to the best size. This algorithm greatly reduces the randomness of neural network recognition, the generated 3D human pose is more stable, greatly reduces the problem of joint jitter offset, and the effect is more smooth and stable.