Vision Based Detection and Identification of Smoke Emitting Vehicles Using Traffic Surveillance

Abstract: The rapid increase in urban vehicular traffic has made real-time monitoring of exhaust emissions a critical priority for environmental regulation. Traditional emission checks rely on stationary, hardware-based testing, which cannot capture dynamic, on-road violations. This paper proposes a non-invasive, vision-based framework designed for integration with existing traffic surveillance infrastructure to automatically detect and identify smoke-emitting vehicles. The system utilizes a multi-layered architecture: an adaptive spatial pre-processing module to establish road horizons, a YOLOv8-driven instance segmentation model for vehicle and plume isolation, and a K-Means colorimetric engine to analyze the severity and type of emission. By dynamically filtering surveillance feeds, the system precisely segments exhaust plumes and autonomously categorizes them (e.g., Black Smoke indicating fuel faults, Blue Smoke indicating oil burning). Processed through an interactive web-based dashboard, this framework provides high-fidelity visual tracking and automated alerts, offering a scalable, software-driven solution for Intelligent Transportation Systems (ITS) to enforce emission standards in real-time.

Keywords: Traffic Surveillance, Intelligent Transportation Systems (ITS), YOLOv8, Instance Segmentation, K- Means Clustering, Environmental Monitoring, Vehicle Emissions.