Real-time rapid accident detection for optimizing road safety in Bangladesh

Area of Science:

• Computer Vision
• Artificial Intelligence
• Traffic Safety Engineering

Background:

• Road traffic accidents in Dhaka result in severe fatalities and economic losses.
• Existing emergency response systems require upgrades for efficiency.

Purpose of the Study:

• To propose an advanced car accident detection system using YOLOv11 for real-time detection.
• To compare the performance of YOLOv9, YOLOv10, and YOLOv11 for accurate accident detection.
• To enhance emergency response systems and improve road safety.

Main Methods:

• Utilized a dataset of 9000 labeled images for training object detection models.
• Employed state-of-the-art object detection techniques including Intersection over Union (IoU) and Non-Maximum Suppression (NMS).
• Implemented and compared YOLOv9, YOLOv10, and YOLOv11 models for accident detection.

Main Results:

• YOLOv11 achieved a Recall of 0.8249, Precision of 1.0000, and mean Average Precision (mAP) of 0.9940 at a 50% IoU threshold.
• The system demonstrated low latency, processing frames in 19.93 ms on a GPU, making it suitable for real-time applications.
• Comparative analysis showed YOLOv11's superior accuracy in detecting and classifying on-road accidents.

Conclusions:

• The YOLOv11 model offers highly efficient and accurate real-time car accident detection.
• AI-driven systems show significant potential for improving road safety and emergency response times.
• Integration of such systems can reduce accident-related fatalities and societal impact.