Human Body Attitude Estimation Based on Gradient Feature of Depth Images
XU Yuefen · Jisuanji gongcheng · 2015
Because of the diversity of human body pose occlusion and self-occlusion in human pose estimation,the accuracy of the system is not strong and the operation efficiency is low.This paper proposes a feature extraction method base on Gradient of Depth(GoD)for feature representation in human body posture estimation.This method firstly calculates each pixel's gradient value in the horizontal direction and vertical direction by using the image's depth information.Then the method calculates the difference between each pixel and its neighborhood pixels and obtains a 4D feature.At the same time,the method optimizes the random forests then estimates the human body posture in the image.The method has a significant improvement of human body posture estimation robustness and the accuracy in the process of single frame image compared with the optimized random forest decision-making method,and improves the test efficiency by reducing the accuracy rate of 0.1%.