AuraPose: Accurate Human Pose Detection and Behavior Recognition via Enhanced OpenPose with Angular Measurement

The rapid progress in computer vision and deep learning technology has positioned human posture detection and estimation as a hot spot research area with sustained development. Nevertheless, prevalent detection technologies frequently encounter issues related to accuracy and robustness which hinders their further practical adoption. In this paper, an enhanced OpenPose algorithm incorporates angular measure-ment is proposed to tackle above challenges. Firstly, OpenPose is utilized to detect and locate key points of human body. Subsequently, the connections between these points according to the specified posture are optimization. Finally, ultimate angles are measured to deliver the accurate pose detection and behavior recognition. Qualitative analysis on recorded videos illustrate the applicability and robustness of proposed method. Furthermore, quantitative results on COCO2017 dataset compared with the state-of-the-art methods also show competitive performance of proposed method, in which our AR, AP and mAP reached 89.2%, 79.8% and 81.8%, respectively. Through meticulous optimization, our method has the potential to further enhance vision-based human posture detection and measurement tech-nologies, particularly within the domains of medical, biomedical and healthcare systems.