Body Posture Detection Based on OpenPose and 3D Reconstruction

3D reconstruction is of great significance in our daily life. Application such as Computed Tomography(CT) in medical. OpenPose is an open source algorithm in Realtime Multi-Person 2D Pose Estimation. It is characterize by bottom-up and faster recognition speed. The most common use is to take a person of pictures to identify. It is tedious to take pictures of a person multiple times in all directions when people need to obtain human full body posture information. In order to solve this problem, our team spins around the person to record a video. The video will be segmented to obtain a two-dimension image dataset. The blur image in the image dataset will be sharpen though DeblurGAN-V2 algorithm. 3D human model will be reconstructed through the usage of openMVG(Open Multiple View Geometry) combines with openMVS(Open Multi-view Stereo). OpenPose human posture recognition algorithm will be used on the human model so as to solve the tedious problem above. The information obtained is relatively more comprehensive at the same time.