Poseaware Fall Net: Fall Detection Using Human Pose Estimation

Falls among the elderly remain a significant health concern, necessitating accurate and real-time detection systems. This research introduces PoseAware FallNet, a hybrid fall detection framework combining lightweight heuristic-based filtering with deep learning verification using pose estimation. The system utilizes MediaPipe for initial fast keypoint extraction and applies a YOLOv7-w6-pose model for secondary validation, ensuring precision while maintaining low latency. Video data is processed in a temporal window surrounding potential fall incidents, allowing contextual analysis for reliable classification. The proposed PoseAware-FallNet outperforms existing methods, achieving 94.20 % accuracy on the URFD dataset and 89.90 % on the Le2i dataset. Its multi-stage pipeline enables scalable deployment in real-world environments such as smart homes and elderly care centers. The system runs efficiently on edge-compatible hardware and supports real-time alerts. Future enhancements include adaptive thresholds, multiperson tracking, and predictive analysis. This work demonstrates the potential of integrating <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$A I$</tex> and medical insight for robust fall detection.