AI-Based Smart Traffic Monitoring System

Manual enforcement of traffic rules in Indian cities is slow, inconsistent and difficult to scale, and most automated systems treat every offence identically regardless of the offender’s history. This paper presents the design and prototype implementation of the AI-Based Smart Traffic Monitoring System, built with Flask and OpenCV. The system registers five classes of camera-detectable violations (illegal parking, red-light jumping, riding without a helmet, wrong-way driving and triple riding) from simulated multi-camera feeds, uploaded images or video, and browser camera snapshots; stores evidence with each complaint; resolves owner details through a simulated RTO lookup; and uses a rule-based Violation Predictor to compute a Rule Breach Rate (15–98%) from prior violations and violation severity. The predictor then decides automatically between an official Warning for first-time offenders and a Direct E-Challan whose fine is escalated by 25% (third offence) and 50% (fourth and later offences), removing operator ambiguity at the review stage. A browser-side offline queue with idempotent retry prevents loss of evidence during server outages. Testing of the prototype showed that the decision logic behaves deterministically across all nine supported violation types, that image analysis takes 3–30 ms per 800×420 image on a CPU, and that the live scene renderer sustains the 45 ms stream interval comfortably. The detection layer is simulated and has not been validated on real CCTV data; this limitation, together with a weakness we found in the thumbnail-based demo matcher, is discussed along with the steps required for real deployment.