Lightweight Deep Neural Network Approach for Parking Violation Detection

Abstract

Routine patrolling and inspection of parking violations is a time-consuming and labour-intensive process. As such, a lightweight deep neural network approach is developed to automate parking violation detection in outdoor parking areas. An IP camera is utilized to continuously capture outdoor parking image covering multiple illegal parking regions and feed it to Raspberry Pi. Detection is performed on the Raspberry Pi by fusing the lightweight image classification model with sliding window search program to locate illegally parked vehicles. The system is able to detect double parking violations and vehicles that parked illegally in unmarked area. Multithreading processing is employed to speed up the detection process. An Android-based smartphone application known as the Enforcer App is developed to translate the detection results stored in server into graphical user interface. The application displays live parking violation information at parking areas as well as the position of each illegally parked vehicle to ease parking enforcement. An initial prototype was implemented at an outdoor parking lot of Multimedia University, Malaysia to study its detection performance. Experimental results demonstrate high reliability and robustness of the proposed system with no missed detection and 98.7% precision rate. The parking violation detection in three illegal parking regions are completed within a minimum time of 3.46 seconds.