Efficient Multi-person Hierarchical 3D Pose Estimation for Autonomous Driving

摘要

With increasing need of analyzing human poses for autonomous driving, multi-person 3D pose estimation using monocular moving camera in real world scenarios is of great concern. Existing 3D human pose estimation features either large scale training data, or high computation complexity due to the high degrees of freedom in 3D human poses. We propose a novel scheme to hierarchically estimate 3D human poses in natural videos by static or moving cameras in an efficient fashion. Our method does not need 3D training data. We formulate torso estimation into a Perspective N Point (PNP) problem, formulate limb pose estimation into an optimization problem, and structure the high dimensional poses to address the challenge efficiently. Experiments show good performance and high efficiency of multi-person 3D pose estimation on real world street scenario videos, resulting in great new opportunities to understand and predict human behaviors for autonomous driving.