Automated Drowning Detection

This repository presents a comprehensive solution for automated drowning detection leveraging deep learning models and advanced methodologies. Our work focuses on revolutionizing drowning incident response systems by swiftly identifying distress situations in aquatic environments.

Methodology Highlights

Our approach involved rigorous model refinement to achieve accurate drowning detection:

• Initial Model Training: We initially trained the YOLO v8 architecture on a dataset but encountered suboptimal detection results.
• Dataset Analysis: Recognizing the limitations, we integrated a secondary dataset, Team Burraq via Roboflow Universe, to augment our model's understanding of varied drowning scenarios.
• Fine-Tuning for Enhanced Precision: To address the limitations observed in initial detection, we subjected the previously trained model to a further 20 epochs of fine-tuning on the new dataset. This refined training significantly improved detection accuracy and sensitivity, evident in the detection images included.

Key Features

• Dataset Collection and Cleaning: Curated diverse datasets from reputable sources, ensuring comprehensive coverage of drowning scenarios.
• Model Selection and Fine-Tuning: Employed YOLO v8 architecture, fine-tuning it exclusively for drowning instances to enhance accuracy and sensitivity.
• Integration and Refinement: Integrated secondary datasets to further refine the model's understanding of varied drowning scenarios.
• Code Implementation: Utilized Ultralytics library for seamless model loading and image analysis, facilitating efficient drowning instance detection.
• Coordinate Calculation Methodology: Incorporated coordinate calculation to precisely determine spatial coordinates of detected objects, enhancing object localization.
• Integration with Real-Life Pool Cameras and Depth Estimation Sensors: Worked towards integrating the detection system with pool cameras and depth estimation sensors for real-time spatial analysis.
• Conclusion and Future Recommendations: Summarized findings, highlighting accuracy improvements and proposed future enhancements for real-time drowning detection systems.

Dataset Used

• First Dataset: Roboflow Universe
• Second Dataset: Team Burraq via Roboflow Universe

Sample Detection Images

Usage

Detailed instructions and code implementation for automated drowning detection are provided in this repo.