A Deep Learning based Illegal Parking Detection Platform

Abstract

Illegal parking is a critical problem in large, growing cities. Currently, the responsibility for detecting illegally parked vehicles has been left to law enforcement, which often requires manual inspection. To improve the efficiency of law enforcement for vehicle parking management, we propose a web-based analytic platform that leverages recent advancements in computer vision. This proposed platform provides an algorithm to improve the performance of detecting vehicle license plates from videos, based on an existing deep learning approach. Also, we provide a method to estimate vehicle parking locations. This platform is applicable for videos of security patrolling. End-users can define restricted zones via a map-based interface and all vehicles located in these areas can be efficiently identified once patrolling videos are received. This system is evaluated by two videos captured in real-world parking lots. The results indicate that the proposed platform can successfully identify vehicle plate numbers and estimate their parking locations to support the management of urban parking infrastructure.