Human Activity Recognition in Construction Industry Using Machine Learning Pose Estimation Technique

In many applications, including measuring physical activity, understanding sign language, and controlling full-body gestures, human position estimation from video is essential. This has the potential to be utilized for activity recognition in civil work. By accurately tracking human body posture from video, the technology can identify and classify different tasks and actions being performed by workers in construction, manufacturing, or other industries. This information can be used to monitor worker productivity, optimize workflow, and identify potential safety hazards. The proposed project is a machine learning (ML) solution for high-fidelity body posture tracking, employing current open source research that also drives the Machine Learning Pose Detection Application programming interface to infer predefined 3D landmarks and background segmentation mask on the entire body from RGB (Red, Green, Blue) video frames. The suggested method in this project achieves real-time performance on the majority of modern mobile phones, desktops/laptops, Python, and even the web, in contrast to current state-of-the-art methodologies, which rely mostly on strong desktop environments for inference.