HUMAN ACTIVITY RECOGNITION WITH OPENCV

Abstract

This paper proposes an intelligent real-time system integrating Human Activity Recognition (HAR) with fire and human detection using advanced computer vision and deep learning techniques. The system employs the YOLO-based object detection model for identifying fire, smoke, and human presence, while HAR is achieved using a hybrid approach combining Convolutional Neural Networks (CNN) for spatial feature extraction and Long Short-Term Memory (LSTM) networks for temporal activity classification. OpenCV is utilized for preprocessing and feature extraction from video streams. The model is trained and evaluated on publicly available datasets such as fire image datasets and human activity datasets (e.g., UCF101). Experimental results demonstrate improved accuracy, real-time performance, and reliable detection under different environmental conditions. The proposed system effectively recognizes human activities such as walking, running, and abnormal behavior during fire incidents, thereby enhancing situational awareness and emergency response. This integrated approach provides a scalable and efficient solution for smart surveillance and disaster management applications.