A practical framework of multi-person 3D human pose estimation with a single RGB camera

Abstract

We propose a practical framework named 'DN-2DPN-3DPN' for multi-person 3D pose estimation with a single RGB camera. Our framework performs three-stages tasks on the input video: our DetectNet(DN) firstly detects the people's bounding box individually for each frame of the video, while our 2DPoseNet(2DPN) estimates the 2D poses for each person in the second stage, and our 3DPoseNet(3DPN) is finally applied to obtain the 3D poses of the people. Experiments validate that our method can achieve state-of-the-art performance for multi-person 3D human pose estimation on the Human3.6M dataset.