A Comprehensive Drowning Detection System Employing Advanced Computer Vision for Enhanced Water Safety

The increasing number of drowning accidents, particularly in swimming pools, highlights the urgent need to improve water safety. However, conventional surveillance methods have shown limited abilities in preventing drowning issues. This research examined previous research in the fields of underwater object detection and drowning behavior as well as explored similar projects and methodologies. A system architecture was proposed to address this issue, which employed computer vision and deep learning technologies and integrated YOLOv3, a powerful object detection algorithm, with real-time video analysis and motion analysis. The impact of parameter modifications was investigated, specifically the Confidence score and Non-Maximum Suppression (NMS) threshold, on the effectiveness and accuracy of drowning detection. The study concluded that a compromise between a Confidence value of 0.5 and an NMS threshold of 0.5 produced optimal outcomes in terms of processing speed and accuracy through iterative experiments. Hence, the proposed drowning detection system has demonstrated improvement in enhancing water safety by reducing response times during drowning incidents.