Modeling and Simulation of Object Detection Algorithm Performance in Resource Constrained Environments

Abstrak

Deploying deep learning-based object detection algorithms on resource-constrained edge devices presents real challenges in the accuracy–inference speed trade-off. Selecting the wrong algorithm can lead to high latency or low accuracy that fails to meet application requirements. This study aims to analyze and compare the performance of several major object detection algorithms YOLOv8 (Nano, Small, Medium variants), SSD MobileNet V1, and EfficientDet Lite (Lite0, Lite2) across various edge device classes through a review and modeling approach based on published benchmark data. The method is a literature-based comparative-analytical study synthesizing inference metrics, COCO mAP, and energy consumption data from 2022–2026 studies. The analysis shows that SSD MobileNet V1 achieves the fastest inference time (22 ms on the Jetson Orin Nano) but the lowest mAP (23.0%), while YOLOv8m delivers the highest accuracy (mAP 50.2%) at greater resource cost. YOLOv8n and EfficientDet Lite0 emerge as balanced candidates with sub-50 ms inference on the Jetson Orin Nano and mAP above 33%. This study concludes that algorithm selection must integrally consider the device, latency requirement, and accuracy tolerance as a combined decision.