Human Pose Estimation: A Comparative Study of HRNet and OpenPose

Abstract

'Human pose estimation' plays a crucial role in computer vision, enabling machines to understand human actions and behaviors. It's widely applicable in various aspects of our lives. Recent advancements in deep learning have led to remarkable progress in this field, surpassing traditional methods.

This project focuses on two prominent deep learning algorithms: HRNet and OpenPose. These algorithms are used to predict the location of key points on the human body, which are then connected based on prior knowledge to reconstruct the human skeleton.

HRNet has gained significant traction due to its effectiveness in tasks like semantic segmentation, object detection, and classification. This project utilizes HRNet to explore its capabilities in human pose estimation.

OpenPose, a pioneering algorithm, leverages convolutional neural networks and supervised learning within the Caffe framework. It excels in tracking various aspects of human pose, including facial expressions, limbs, and fingers, with impressive performance for both single and multi-person detection.

The project concludes by demonstrating practical applications of these algorithms. OpenPose is implemented to achieve limb-driven functionality, while HRNet enables expression-driven capabilities.