Real-Time Road Traffic Condition Detection System Implemented on Limited Resources Microcontroller Using Lite Deep Learning Techniques

Abstract

With the great progress and development witnessed by smart systems, especially Internet of Vehicles systems, they face several obstacles and challenges. The most important of these challenges is to propose an embedded system to detect congested and non-congested roads with vehicles in real-time that are located in the same place without taking another more suitable path implemented on a low-cost, small-sized, resource-limited embedded device that can be placed anywhere in the vehicle using deep learning techniques. The main goal of this work is to introduce a novel low-cost mobile embedded system that can be placed in a vehicle for monitoring and analysing real-time traffic conditions on public roads. The system is based on a resource-limited Arduino Nano Sense 33 BLE rev1 microcontroller with CPU Flash Memory: 1MB and SRAM: 256KB integrated with an OV7675 camera to capture live streaming videos. A customized deep learning model based on Convolution Neural Networks (CNNs) and Mobile Net V2 with transfer learning is employed for image classification, enabling accurate detection and identification of crowded and non-crowded traffic scenarios. The novelty focuses on optimizing the system for real-time crowded detection performance using resource-constrained hardware, making it suitable for deployment in urban environments. Experimental results show that the proposed system achieved an accuracy of 96.9% with inference time of 328ms. Our approach illustrates significant improvements compared to existing solutions in accuracy and efficiency, offering a good tool for smart city traffic management.