Person Fall Detection Using Weakly Supervised Methods

Person falls can result in severe injuries or fatalities. An automatic fall detection system can potentially save lives by promptly alerting others. Existing fall detection methods that employ physical sensors like accelerometers have limitations. Current computer vision-based approaches, trained on simple and unrealistic datasets, also lack effectiveness. Creating a new dataset for traditional supervised learning would require a significant amount of time to annotate. To address this, we adopt weakly supervised methods from Video Anomaly Detection (VAD) and curate a highquality and realistic dataset. Our proposed model, utilizing Multiple Instance Learning, introduces a novel loss function that outperforms state-of-the-art anomaly detection models for fall detection. Furthermore, despite its simplicity, our approach achieves competitive performance compared to the current state of the art in UCF-Crime.