Human Posture Identification and Recognition Using Deep Learning Techniques

Human Pose Estimation is the task of estimating the human joints in a 2D or 3D coordinate plane from an input RGB image. The 2D HPE are of two types: top-down and bottom-up. The 3D Human Pose Estimation gives better understanding of the relative position of the joints of the person and is helpful to understand the positions of close joints. The evaluation metrics and the architectures for 2D HPE are discussed in brief in the paper. The paper focuses on 3D Human Pose Estimation because it provides more information about the depth of the joint and gives better spatial understanding of the joint’s position. The evaluation metrics and various 3D HPE architectures: Simple Baseline, Lifting from the Deep, Camera Distance Aware approach, and BlazePose. A custom in-the-wild dataset was built and was used for evaluating the 3D Human Pose Estimation architectures.