Object-Based Hybrid Deep Learning Technique for Recognition of Sequential Actions

Using different objects or tools to perform activities in a step-by-step manner is common practice various settings, including workplaces, households, and recreational activities. However, this approach can pose several challenges potential hazards if the correct sequence of actions not followed object tool used appropriate sequence; therefore, it must be addressed ensure safety efficiency. These issues have garnered significant attention recent years. Previous research has relied on using body keypoints detect actions, but during activity. As result, lack system identify target being while performing tasks increases risk accidents mishaps process. This study suggests possible solution aforementioned issue by introducing model that both efficient durable. The utilizes video data monitor daily activities, as well involved process, thus enabling real-time feedback alerts enhance productivity. suggested separates overall recognition process into two components. Firstly, advanced BlazePose architecture for estimation, interpolates any undetected wrong-detected landmarks precision posture estimation. After this, features are forwarded long short-term memory network performed Secondly, also employs an enhanced YOLOv4 algorithm detection, accurately course Finally, durable activity been developed, which achieves 95.91% accuracy rate identifying mean average score 97.68% detecting objects, capable processing at 10.47 frames per second.