The Research On Human Post Emotion Based On Depth Images

Human body in video of intelligent monitor has attracted much attention in the field of computer vision. Body pose is a promising topic which is popularly used. The estimate of the body parts has become a significant area of the research. Affected by many factors such as the complexity of human pose, self-occlusion or covered by others, light, etc, there isn’t any ideal solution for pose estimation that is effective and robust.A new method of human pose estimation is proposed based on the depth images which using illumination invariance of depth images and consistency of spatial information.Considering that traditional feature extraction method is inappropriate for human pose estimation, this paper proposes a feature extraction method based on GoD(Gradient of Depth) 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. Experimentally results show that GoD has a big promotion both on accuracy and robustness.For the destination of human body pose estimation is the application in reality, a high efficiency is necessary. On the basis of traditional random forest, this paper reduces the expenditure of time by optimizing random forest which utilizes the weight sequencing of the decision-making tree and reduction of the quality of decision-making tree in ultimate step. Experimental results show that improved random forest has a smaller time cost in test phase and good pose estimation accuracy.In the end, the thesis makes a conclusion of the study and points out some shortages and looks forward to the future.