Central AsianResearch Index

REAL VAQT REJIMIDAGI VIDEOKUZATUV TIZIMLARIDA OBYEKTLARNI ANIQLASH VA TANISHNING NEYRON TARMOQ USULLARI: SIRKULYAR IQTISODIYOT VA RESURS SAMARADORLIGI YONDASHUVI

The article systematized neural network methods for object detection, tracking and face recognition in video surveillance from a circular economy perspective. It showed that published results were not directly comparable because of different datasets and GPUs. An analytical model of end-to-end frame latency and of the number of streams per accelerator was proposed: in a hypothetical example, invoking the detector on every fifth frame raised the number of streams from 3 to 12. Recommendations to restrict face recognition and save hardware were developed. The analysis covered the published results of YOLO, Faster R-CNN, YOLOv7, RT-DETR, DeepSORT, ByteTrack, FaceNet and ArcFace. The trade-off between detection frequency, stream capacity and the delay in detecting a new object in a cascade architecture was shown. The recommendations rely on the legal requirements for processing biometric data and on operator oversight. The proposed model makes it possible to estimate in advance the number of accelerators required when designing a video analytics system.

Сведения

Журнал
Marketing Jurnali
Опубликовано
2026 · С. 281-286
Язык
Узбекский
ISSN
3060-4621
Издатель
"TADBIRKOR VA ISHBILARMON" LLC
Аффилиации
Ichki ishlar vazirligi Tashkiliy departamenti Situatsion-tahliliy markazi smena boshligʻi