نظام الكشف التلقائي عن مخالفات قواعد المرور

Automatic Traffic Rules Violation Detection System

Abstract

Traffic violations cause significant problems such as congestion, accidents, and deaths. It is highly desirable to have an effective automated system to detect and record these violations, thus, to improve traffic regulation enforcement and reduce human intervention. This study aims to detect traffic violations and identify violated vehicle plates in Iraq. As the system detects traffic violations automatically, of three different cases; non-compliance with the traffic signal, exceeding the speed limit, and driving in the opposite direction, it is cost-effective, practical, and powerful. The proposed system uses background subtraction technology to detect moving vehicles and the concept of time and distance difference over which vehicles move to detect violations. The You Only Look Once version 4 (YOLOv4) algorithm is used to identify the license plates (LPs) of violating vehicles with great accuracy. The Convolutional Recurrent Neural Networks (CRNN) were used with Optical Character Recognition (OCR) technology to recognize two types of Iraqi LPs; the first type contains Hindi numbers, while the second form includes Arabic numbers, a future system to be used throughout Iraq. The results achieved from our trial indicate promising system performance, with multiple real-time violation detection and an overall accuracy rate of 98.06% for offenses. On the other hand, the proposed system has excelled in recognizing LPs, achieving a success rate of 94.87% and 98.22% for the first and second types, respectively. It has outperformed similar systems, particularly in terms of accuracy and the capability to detect multiple types of traffic violations simultaneously, thereby establishing its competitive edge.