Primerjalna analiza lahkih modelov za zaznavanje objektov v maloprodajnem videonadzoru na edge napravah

KOVAČEVIĆ, DEJAN (author), Šajn, Luka (mentor)

V diplomski nalogi primerjamo pet lahkih modelov za zaznavanje objektov – YOLOv8n, YOLOv8s, YOLOv5n, NanoDet-Plus in MobileSAM – v kontekstu maloprodajnega videonadzora. Modele ocenimo na podmnožici testne množice COCO 2017, ki zajema 2693 slik z oznakami razreda oseb, z uporabo pred- naučenih uteži na prenosnem računalniku Apple MacBook Pro z M2 Max procesorjem. Primerjamo jih po petih merilih: hitrosti izvajanja (FPS), točnosti (mAP@0,5, mAP@0,5:0,95), povprečnem priklicu (AR@0.5), velikosti modela in računski zahtevnosti (GFLOPs). Rezultati kažejo, da je NanoDet-Plus najhitrejši model (66,0 FPS) z najnižjo računsko zahtevnostjo (2,97 GFLOPs), vendar z nižjo točnostjo, še posebej pri strogem merilu mAP@0,5:0,95 (19,97 %). YOLOv8s dosega najboljšo točnost (mAP@0,5 = 70,26 %), a pri nižji hitrosti (19,6 FPS). Za ce- lostno najboljše razmerje med hitrostjo in točnostjo se izkaže YOLOv8n (30,0 FPS, mAP@0,5 = 64,36 %), ki ponuja dobro ravnovesje za uporabo v malopro- dajnem okolju na napravah z omejenimi viri. YOLOv5n je najkompaktnejši model (1,9 M parametrov, 4,0 MB), primeren za okolja z zelo omejenimi viri pomnilnika. MobileSAM kljub dodani segmentacijski zmogljivosti ni konkurenčen čistim detektorjem v hitrosti izvajanja (2,7 FPS). V nalogi podrobno opišemo arhitekturne razlike med modeli, izzive pri implementaciji enotnega merilnega cevovoda ter analiziramo razmerja med posameznimi metrikami. Ugotovitve so relevantne za načrtovalce sistemov računalniškega vida, ki izbirajo med različnimi lahkimi detektorji za namesti- tev na edge napravah v maloprodajnem okolju.

• Language: Slovenian
• Keywords: računalniški vid, zaznavanje objektov, YOLO, NanoDet, MobileSAM, maloprodaja, videonadzor
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2026
• PID: 20.500.12556/RUL-180740
• COBISS.SI-ID: 275433987
• Publication date in RUL: 16.03.2026

Secondary language

• Language: English
• Title: Comparative Analysis of Lightweight Object Detection Models for Retail Video Surveillance on Edge Devices Description: The goal of this thesis is to compare real-time performance, resource consump-

Abstract

This thesis presents a comparative analysis of five lightweight object detection models – YOLOv8n, YOLOv8s, YOLOv5n, NanoDet-Plus, and MobileSAM – in the context of retail video surveillance. We evaluate the models on a subset of the COCO 2017 dataset comprising 2,693 images with person-class annotations, using pre-trained weights on an Apple MacBook Pro with an M2 Max processor. The comparison covers five metrics: inference speed (FPS), accuracy (mAP@0.5, mAP@0.5:0.95), average recall (AR@0.5), model size, and computational complexity (GFLOPs). Our results show that NanoDet-Plus is the fastest model (66.0 FPS) with the lowest computational cost (2.97 GFLOPs), though at the expense of lower accuracy, particularly at the strict mAP@0.5:0.95 metric (19.97 %). YOLOv8s achieves the best accuracy (mAP@0.5 = 70.26 %) but at a reduced speed of 19.6 FPS. For the best overall trade-off between speed and accuracy, YOLOv8n (30.0 FPS, mAP@0.5 = 64.36 %) proves most suitable for real- time retail deployment on resource-constrained devices. YOLOv5n offers the smallest footprint (1.9M parameters, 4.0 MB), while MobileSAM, despite providing additional segmentation capabilities, is not competitive in inference speed (2.7 FPS). We provide detailed descriptions of architectural differences between the models, discuss the challenges of implementing a unified benchmarking pipeline, and analyze the trade-offs between individual metrics. Our findings are relevant for system designers choosing between lightweight detectors for deployment on edge devices in retail environments.

• Keywords: computer vision, object detection, YOLO, NanoDet, MobileSAM, retail, video surveillance