A Hybrid Deep Learning and IoT Architecture for Automated No-Parking Enforcement: Performance Analysis and Implementation

Abstract

Illegal parking is a critical urban challenge, accounting for an estimated 20-30% of traffic congestion and contributing to significant economic and environmental costs [1, 2]. This paper presents a robust, cost-effective Internet of Things (IoT) system designed for autonomous detection of no-parking violations and instantaneous owner notification via SMS. The system integrates an ultrasonic sensor (HC-SR04) for vehicle detection, a Raspberry Pi 4 with a camera module, the Firebase cloud platform, and the Fast2SMS API. The core innovation lies in the implementation and statistical validation of a hybrid image processing pipeline. This pipeline combines traditional computer vision techniques for license plate localization with a deep learning-based Convolutional Recurrent Neural Network (CRNN) for Optical Character Recognition (OCR), specifically using the EasyOCR engine [3]. Rigorous testing over 100 detection events yielded an overall system accuracy of 88%, with a mean processing time of 4.2 seconds per event. The proposed solution offers municipalities a scalable, high-efficacy framework for intelligent traffic management, moving decisively towards the realization of smart cities [4].