Pose Estimation on 3-D Models Using ConvNets

Abstract

Pose estimation of various human activities has seen tremendous progress over the past few decades due to the advancement of computer vision technology. However, the current status of pose estimation is quite stagnated when seen in context to human behaviour analysis by using the pose. In this paper, pose estimation has made significant advancement in terms of dexterity and ingenuity, which is sufficient for human pose estimation. This work includes a large dataset of various human activities. The collected images cover a wide range of human activities that include various activities in various postures in different angles of viewpoints. We have annotated all images for various joint locations such as head, elbow, hands, and eyes via CSV data file. The pose analysis of this work will provide various ongoing research insights about the use of computer vision and Convolution neural networks (ConvNets) for practical purposes and would further instigate this technique as an alternative for VFX in the film industry.