Leveraging YOLO for Real-Time Human Detection and Pose Estimation in Live Stream Environments

This proposed framework presents a real-time object detection and monitoring system using YOLO (You Only Look Once) models and OpenCV for video processing and analysis. Combining classification and localization, the system employs two YOLOv8 models for human pose detection and object segmentation. It processes live video streams and integrates features like thermal vision simulation, flipped video views, and dynamic location updates to mimic drone behaviour. Annotated results are displayed via a Flask web interface for real-time monitoring. The dataset is labeled with bounding boxes and split into training, validation, and test sets to ensure robust performance. YOLO's single-stage detection pipeline achieves high-speed, accurate inference. Detection results and warnings are logged into a CSV file, with alerts for high object counts. Trained model weights in. pt files enable efficient transfer learning and inference. This modular system integrates object detection, data logging, and web visualization, demonstrating applications in surveillance, autonomous systems, and human activity recognition, with flexibility for diverse real-time scenarios.