Deep Learning Based Traffic Accident Detection in Smart Transportation: A Machine Vision-Based Approach

This paper provides a deep learning-based method for traffic accident identification in smart transportation that is based on machine vision and uses the YOLOv8 architecture. The goal of the research is to precisely identify and localize accidentrelated factors to improve safety protocols and system efficiency in transportation. The YOLOv8 model performed exceptionally well using deep learning approaches, producing a mean Average Precision (mAP) of 94.4%, Precision of 91.6%, and Recall of 92.3%. The study focused on the testing and inference phases and thoroughly assessed the model's capabilities. High identification rates throughout testing across multiple scenarios showed how well the program could recognize accidents, including car crashes and non-accident scenes. The model's accuracy and dependability were highlighted by its capacity to identify non-accident scenarios without producing false positives. These encouraging results underline the YOLOv8 architecture's preparedness for implementation and show its potential to raise efficiency and safety standards in smart transportation networks greatly. This work represents a significant step forward in the field of machine vision-based accident detection and suggests future directions for improving real-time, accurate accident identification for more secure and effective transportation systems.