Real-Time Object Detection Algorithm Performance on Edge Computing Devices

Distracted driving remains a major cause of traffic accidents, highlighting the critical need for driver assistance systems. Edge computing offers a promising solution by enabling real-time data processing at the source, reducing power consumption, and improving portability. This study investigates the performance of YOLOv5, a state-of-the-art object detection algorithm, for real-time traffic light detection on three popular edge devices with different processor architectures: Jetson Nano, Raspberry Pi 4, and Coral TPU Dev Board. We evaluate each device based on its inference time, and power consumption, providing a comprehensive comparison for deploying a small, power-efficient driving assistant system. Our results demonstrate the potential of YOLOv5 on-edge devices for real-time traffic light detection, offering valuable insights for researchers and developers in intelligent transnortation systems.