Real-Time Drowning Detection and Marine Hazard Identification
This project involved annotating video and image datasets for a real-time drowning detection system and marine hazard identification. Specific tasks included creating bounding boxes and polygons to identify swimmers, hazardous marine life (e.g., sharks), and environmental conditions (e.g., rip currents). The segmentation task further refined the delineation of swimmer positions, ensuring accurate detection in dynamic water environments. The dataset included thousands of frames requiring precise annotation to handle challenging scenarios such as occlusions, varying light conditions, and complex underwater movements. To ensure quality, a multi-step validation process was implemented, involving peer review and automated checks for annotation consistency. The labeled dataset contributed to developing an AI system with high detection accuracy, significantly improving lifeguard response times.