Person Posture Estimation based on Pose Angular Feature and Region Keypoints Detector Network

Computer vision and digital image recognition researchers face a difficult problem with the video analysis-based Person Posture Estimation (PPE) system. Several surveillance PPE applications rely on well-designed learning algorithms for their performance. The complexities of human movements, such as motion, posture, scale variation, viewpoint variation, and occlusions, are among the challenging study concerns of the video-based PPE system. The computer vision-based key points pose an estimation to provide a competitive answer to these challenges. In this paper, the PPE system utilized this pose estimation method to extract multi-scale features and calculate daily activities of pose angular features. Experiments are conducted on our dataset and human daily activity dataset at Kaggle. The mean average accuracy of 95.1% is achieved by using the proposed PPE system that can run on lightweight, low computation-intensive devices.