Traffic Accident Detection Analysis Using YOLOv9 Algorithm

Traffic accidents are a leading cause of death and serious injury worldwide, resulting in approximately 1.35 million fatalities each year. Early detection and rapid response to traffic accidents are deemed crucial for reducing fatality rates and other negative impacts. This study analyzes the application of the YOLOv8 algorithm in detecting traffic accidents, leveraging an AI-based system to enhance detection efficiency and accuracy. Using a dataset of accident images collected through web scraping and open-source platforms, three training scenarios were conducted: YOLOv8 for 100 epochs, YOLOv9 for 100 epochs, and YOLOv9 for 150 epochs. Evaluation results show that the YOLOv9 model with Generalized Efficient Layer Aggregation Network (GELAN) architecture achieved a precision of 94.2% after 100 epochs and 93.4% after 150 epochs, surpassing the performance of YOLOv8, which reached 89.8%. The YOLOv9 model also demonstrated superior and consistent recall and mean average precision (mAP50-95). These findings indicate that the YOLOv9 model with the GELAN architecture can significantly assist in real-time accident detection automation, improving the speed and accuracy of post-accident responses, and contributing to the reduction of fatalities and negative impacts of traffic accidents.