Traffic enforcement and surveillance constitute critical components of Intelligent Transportation Systems (ITS). Contemporary approaches leverage computer vision, deep learning, and edge computing to automate traffic violation detection, incident detection, and enforcement. However, most systems address isolated aspects such as violation detection, accident recognition, or adaptive traffic control without unifying these into a single real-time framework with automated fine allotment and emergency alerts. This survey critically evaluates state-of-the-art vision-based models for traffic monitoring and accident detection, compares methodologies and performance metrics, and identifies key gaps. Based on this, we propose a vision-centric integrated framework that delivers real-time violation detection, fine generation and notification, and accident detection with alerting suitable for smart city deployment. This work synthesizes insights from multi-disciplinary research, highlighting strengths and limitations of current vision-based traffic systems and charting a path for future research.
JJ et al. (Sun,) studied this question.