A Deep Learning Based Method For 3D Human Pose Estimation From 2D Fisheye Images

We propose a deep learning based method to directly estimate the human joint positions in 3D space from 2D fisheye images captured in an egocentric manner. The core of our method is a novel network architecture based on Inception-v3 [4], featuring the asymmtric convolutional filter size, the long short-term memory module, and the anthropomorphic weights on the training loss. We demonstrate our method outperform state-of-the-art method under different tasks. Our method can be helpful to develop useful deep learning network for human-machine interaction and VR/AR applications.