Dynamic Pose Estimation for Drill Activites Using Deep Learning

Over the past 20 years, Human Activity Recognition (HAR) has gained popularity as a research area. Comprehending human behavior in images has numerous applications, such as pose estimation and scene recognition, and provides valuable information for a wide range of computer vision problems. There are several approaches available for activity recognition; each approach has benefits and drawbacks. Even with extensive research, identifying activity remains a difficult and complex task. This paper proposes a method that uses an individual's pose skeleton in images to recognize and classify human activities. The two sections of the work have been the estimation of poses for a particular person and pose-based activity categorization. Pose estimation involves identifying eighteen body joints along with significant points. Here, Open Pose library is used for pose estimation work; additionally, OpenCV is also used to perform image processing and provide wide range of features like object detection, face recognition and tracking. In this paper, pose detection for drill activity is being worked on as a component of Human Activity Recognition (HAR). The project's data set was made from scratch and includes various positions such as Saavdhan, Vishram, Salute, Dahine Mud, Bahine Mud, and Pichhe Mud. Furthermore, Pyxl libraries were used along with open CV and Open Pose.