Deep learning-based route an‎d behavior prediction based on traffic signs in self-driving vehicles

10/19/2026 12:00:00 AM

Self-driving cars, as one of the most advanced innovations in the automotive industry an‎d intelligent transportation technologies, are transforming the global transportation lan‎dscape. The use of deep learning in autonomous vehicles is a major development in this technology, enabling the real-time processing of vast amounts of data an‎d complex decision-making. However, this approach faces several challenges, such as computational complexity an‎d the need for powerful an‎d expensive hardware. TinyML, as an emerging an‎d important branch of machine learning, focuses on implementing deep learning models on resource-constrained devices like microcontrollers, smartphones, an‎d other embedded systems. In this paper, a cost-effective model car is designed an‎d implemented, utilizing TensorFlow Lite as a key tool to deploy optimized an‎d compact deep learning models on such devices. This model car identifies the driving path using the MobileNet network combined with the Haar Cascade algorithm, an‎d it detects an‎d acts upon traffic signs with high accuracy. These processes are performed in real-time on a small an‎d inexpensive Raspberry Pi computer. The final system achieved 98.82% accuracy in lane detection while maintaining its real-time performance with a total execution time of 180 milliseconds an‎d a power consumption of 4.9 watts. These results demonstrate that using lightweight deep learning models within the TinyML framework can be a practical an‎d low-cost solution for developing autonomous vehicles on embedded hardware platforms.

خودرو خودران، يادگيري عميق، تشخيص مسير، پردازش ت وير، سيستمهاي نهفته، ماشين كوچكمعياس، بعهبند ي عالئم راهنمايي رانندگي .

Self-driving car, Deep learning, Lane detection, Image processing, Embedded systems, TinyML, Traffic sign classification

Mohammadreza fatan

Mohammad Hassan shojaee-fard