Hardware-Aware Benchmarking of Real-Time Object Detection: From Raspberry Pi to High-End GPUs

<span lang="EN-US" style="font-size: 12pt; line-height: 17.12px; font-family: &quot;Times New Roman&quot;, serif;">This study investigates the performance variations of AI-based object detection algorithms across different hardware platforms. Two prominent models—YOLOv5 and Faster R-CNN—were extensively tested on three Raspberry Pi models (3B+, 4B, and 5) and five GPU-supported laptops with varying computational capacities. The research focuses on evaluating performance through critical metrics such as processing speed, memory usage, latency, and frames per second (FPS). Results indicate that Raspberry Pi 3B+ was unable to run the tested models due to its 1 GB RAM limitation, while Pi 4 and Pi 5 provided more stable and efficient outputs. High-end GPU-enabled systems excelled in speed and FPS, particularly for complex tasks, albeit at higher energy consumption. Notably, under certain conditions, mid-range systems like those with MX450 GPUs demonstrated competitive performance with higher-end setups. These findings underscore the importance of appropriate hardware selection and provide actionable insights for researchers designing systems for real-time object detection tasks. The study concludes with recommendations for optimizing AI deployment across heterogeneous environments.</span><span style="font-size: medium;"></span>