Automated Detection and License Plate Recognition of Motorcycles on Sidewalks Using Deep Learning

Abstract

Motorcycle riding on sidewalks is a common traffic violation in urban areas and poses significant risks to pedestrian safety. However, existing approaches remain limited in processing surveillance images under real-world conditions, particularly those captured from top-view CCTV cameras with constrained resolution. In addition, motorcycle license plates are significantly smaller than automobile plates, resulting in fewer pixels available for character recognition and increasing the difficulty of optical character recognition (OCR). This study proposes an automated system for detecting motorcycles on sidewalks and recognizing license plates from CCTV images. The system employs deep learning techniques for object detection and OCR for extracting textual information from license plates, with a focus on real-world deployment using top-view camera perspectives. The proposed system consists of three main stages: motorcycle detection, license plate detection, and character recognition. A YOLO-based model is trained on annotated datasets for detection tasks, and an image enhancement process is applied prior to OCR to improve recognition performance. Experimental results show that the system achieves average confidence scores of 0.80 for motorcycle detection and 0.84 for license plate detection. In addition, OCR accuracy improves from 32% to 46% after image enhancement. The results demonstrate that the proposed system can effectively operate in real-world environments for detecting illegal motorcycle behavior on sidewalks and has the potential to be further developed into an automated traffic law enforcement support system.