A Multi-Stage Framework for Vehicle Emission Detection and Automated License Plate Recognition (ALPR)

Abstract: Environmental pollution and public health issues due to vehicular emissions are one of the important sources of air pollution in cities. However, traditional vehicle monitoring systems are mostly based on periodic inspections and manual enforcement measures, and they are not always efficient and can't deliver real-time monitoring. An integrated framework of Artificial Intelligence (AI) based real-time vehicle emission monitoring and automated vehicle identification system is presented in this paper. The proposed framework includes machine learning, computer vision, video processing and optical character recognition (OCR) in a conditional execution architecture. First, a Random Forest (RF) classifier is used to classify the driving conditions as polluting or non-polluting on the basis of parameters related to emissions. Computer vision modules are only triggered when a vehicle is determined as a polluter, which helps to lessen the computational load. Video Processing Methods are used to extract frames, Best-frame selection algorithm is used to select the best frame in the video based on the sharpness and brightness of the frame, YOLOv8 is used to detect the number plate of the vehicle, and EasyOCR is used to detect the registration number of the vehicle. Experimental results are shown to obtain 100% classification accuracy for the prediction of pollution and 99.46% mAP@50 for NPD. This framework is a smart, scalable, efficient and intelligent solution for smart city pollution monitoring and automatic regulation assistance.

Keywords: Vehicle Pollution Monitoring, Random Forest, YOLOv8, EasyOCR, Number Plate Recognition, Machine Learning, Computer Vision, Intelligent Transportation Systems.