An Improved 3D Human Pose Estimation Model Based on Temporal Convolution with Gaussian Error Linear Units
PubDate: Aug 2022
Teams: Dalian University & Dalian University of Technology
Writers: Jian Kang; Rui Liu; Yijing Li; Qian Liu; Pengfei Wang; Qiang Zhang; Dongsheng Zhou
Abstract
With the advancement of image sensing technology, estimating 3D human poses from monocular video has become a hot research topic in computer vision. 3D human pose estimation is an essential prerequisite for subsequent action analysis and understanding. It has a wide range of applications, such as intel-ligent transportation, human-computer interaction, and medical rehabilitation. Currently, some methods for 3D human pose estimation in monocular video employ temporal convolutional network to extract inter-frame feature relationships, but the majority of them suffer from insufficient inter-frame feature relationship extractions. In this paper, we decompose the 3D joint location regression into bone direction and bone length, we propose a temporal convolutional network incorporating Gaussian error linear units (TCG) to solve bone direction. It enables more inter-frame features to be captured, allowing the feature relationships between data to be fully utilized. And we use kinematic structural information to sovle bone lenght which enhance the use of intra-frame joint features. The proposed method has extensively experimented on the public benchmark dataset Human3.6M. The quantitative and qualitative evaluation results show that the proposed method can achieve more accurate 3D human pose estimations.
