Optimized Obstacle Detection on Edge Devices

This paper investigates the performance of recorded high-resolution video processing on edge devices, specifically the Jetson Nano and Jetson Xavier, in comparison to a high-performance computing machine equipped with a Nvidia GPU. Our study aims to develop an efficient approach to optimize processing speed, throughput, power consumption, resource utilization, and reliability. We present a novel method for distributing workloads and tasks in manageable chunks to maximize the capabilities of edge devices. Through comprehensive experiments, we compare the performance of CPU and GPU on these edge devices and a powerful computing system. Our results demonstrate that with appropriate optimization, edge devices can achieve high-accuracy predictions with minimal hardware resources, offering a cost-effective and efficient solution for real-time video processing in various applications.