Making AI Accessible: Evaluating YOLOv5n for Real-Time Object Detection in Low-Light Environments on CPU-Only Devices

Abstract—Aspiring AI practitioners in low-resource settings often lack access to high-end hardware (GPUs) and optimal lightingconditions, creating a significant barrier to the democratizationof AI.This study investigates the performance of YOLOv5n, alightweight object detection model, as a viable solution for theseenvironments. We evaluated the model using only CPU-powereddevices, including virtual machines and a MacBook Pro, understandard and simulated low-light conditions. Performance wasquantitatively assessed on multiple metrics, including model loadtime, inference time, CPU/RAM usage, and detection accuracy.The results indicated that while YOLOv5n demonstratedstrong speed and high efficiency in standard lighting, its overalleffectiveness, particularly its processing speed and detectionaccuracy, significantly degraded in low-light scenarios, especiallyon the most limited hardware. However, the MacBook Proconfiguration successfully broke the 10 FPS barrier, establishinga performance ceiling for CPU-only deployment.The general speed and low resource consumption of theYOLOv5n model, even under challenging conditions, confirmits potential for practical deployment in low resource settings.This work provides critical benchmark data necessary to makeinformed hardware and environmental decisions when deployinglightweight AI solutions worldwide.Index Terms—Real-time object detection, YOLOv5n,Lightweight deep learning models, Low-light environments,CPU-only inference, Resource-constrained devices, Accessibleartificial intelligence