Vision-based human pose estimation for learning behaviour analysis: a review
Zhao Li, Dazhen Shen, Tingting Liu, Zhuowei Wang, Weixin Li, Hai Liu, Junya Si, Jiawen Li, Zixin Li · 2023
In this article, we review the vision-based human pose estimation (HPE) methods in the past decades. Human pose estimation is the connection of all detected human body keypoints according to their interdependencies to form a structure similar to the skeleton. Furthermore, human posture is utilized in a wide variety of computer vision tasks, for instance, learning, sports, medical and human-computer interaction. Due to partial occlusion, noise problems, scale changes, and complex backgrounds, HPE remains an important issue for sustainable discussion and development. To obtain human posture from signals or perform behavioural analysis on the human body, HPE uses a variety of methods, including various transformer models and CNN models. Our methods are among them which in CNN module. Specifically, our EHPE and EDE achieve 79.1AP and 78.5AP on COCO dataset.