A Comprehensive Study on Human Pose Estimation

Abstract

The Human pose estimation is the process of digging out predefined body parts called joints from the image/frame inputs. The human body comprises of various bones, muscles and joints with their own movement angles. These are represented with different models. The keypoint representation is one among them. The pose estimation involves locating, grouping and tracing keypoints. Each of the keypoint represents various postures. The neural network the base model for pose estimation tracks these postures The human manners can be better focussed with a proper real-time fine -grained human pose tracker. The single or multiple people in the scene from the monocular or stero cameras are detected based on poses. The pose estimation is an arising research topic due to the wide range of functions in human-robot interaction, gaming, sports performance analysis. The paper discusses about fundamentals of 2D,3D poses estimations, its evaluation metrics, dataset. The paper is over and done with challenges and research directions