Human pose recognition via adaptive distribution encoding for action perception in the self-regulated learning process

Abstract

Human pose recognition (HPR) under infrared imaging is considered a significant role for action perception in the self-regulated learning process. In this article, we propose a novel elliptical distribution encoding (EDE) method for understanding the human behavior through body poses under infrared imaging. Firstly, to reveal the relationship between the adjacent joint points, the elliptical Gaussian coordinate coding is proposed for the human skeleton direction description. Then, the KL divergence is introduced to measure the different between the predicated heatmap and the groundtruth one. Finally, the EDE method is executed on the infrared human pose images in the classroom. The proposed pose recognition is performed in the Python tool. Experimental results are tested on two standard datasets like IRPSRL and MS COCO, and it provides very well classification accuracy. The performances of the proposed method are higher in all performances as related to other existing schemes. The developed EDE method will promote the development of computer-assisted technology in the self-regulated learning process. Our code is publicly available at https://github.com/hai-LIU/EDE.