DNN-Powered Pose Detection for Action Detection

This paper presents a real-time human action recognition system based on pose detection using deep neural networks. A pre-trained TensorFlow model integrated with OpenCV’s DNN module is used to extract keypoints from the human body, forming a skeletal structure for action classification. Rule-based logic, derived from the spatial relationships among joints, enables recognition of actions such as standing, sitting, waving, and hands up. The system processes images, video streams, and webcam feeds without requiring additional training, making it lightweight and efficient. Its modular design allows easy extension, making it suitable for surveillance, gesture control, and human-computer interaction.