Evaluating the Assessment of CNN and YOLO Based Models for Detecting Road Accidents

Infrastructure development and improved socioeconomic status have incorporated various challenges for cutting-edge transit networks. The increasing number of vehicles on the road has led to a higher likelihood of accidents, underscoring the importance of road safety. Adopting strong safety protocols can substantially lower the risk of accidents, and early detection of incidents can help minimize traffic congestion and prevent additional collisions. To address these issues and enhance overall system performance, the study presents a comparison between two modeling approaches: Convolutional Neural Network (CNN) and You Only Look Once (YOLOv8). The CNN architecture is widely used for prioritizing accuracy, and the YOLO architecture is optimized for real-time speed with slightly trimmed accuracy. They both can be used in object detection. This paper examines the performance of CNN and YOLOv8 for accident detection. Both demonstrate advanced deep learning models. The observations show that CNN offers high accuracy and balanced precisionrecall capabilities, whereas YOLOv8 outperforms CNN in terms of speed. The CNN-based accident detection model shows a remarkable performance with an impressive F1 score of 0.929. In comparison, the YOLOv8 model is optimized for speed, while processing data in real-time achieves a peak F1 score of 0.56, given a low confidence threshold of 0.05, suggesting its best performance occurs when predictions are made conservatively.