Automatic Detection of Parking Violation and Capture of License Plate

Illegal parking is a ubiquitous problem faced by urban cities, posing potential traffic impediments and safety risks to other road users. Surveillance video captured is stored for post-event forensics and often requires manual inspection to detect the violating vehicles. Automated detection techniques include foreground and background segmentation which is less robust to environmental factors as well as the more robust Single Shot MultiBox Detector (SSD) and its variants. This paper proposes a fully automated pipeline to perform end-to-end illegal parking detection. Robust and fast vehicle detection is achieved using a deep learning-based object detection algorithm, You Only Look Once Version 3 (YOLOv3). Movement tracking uses template matching and Intersection over Union (IoU) calculations to track the stationary time of the vehicle-in-violation with built-in mechanisms for error tolerance. OpenALPR is used for extraction of the license plate. Empirical results show a high accuracy of vehicle detection and movement tracking under varying environmental conditions and very good performance in license plate capture.