ALGORITHMS FOR AUTOMATIC OBJECT RECOGNITION IN VIDEO DATA
Keywords:
object detection, video analysis, YOLO, Faster R-CNN, EfficientDet, deep learning, real-time systems, YOLOv8Abstract
Real-time object detection and recognition in video streams is one of the most important tasks in modern computer vision. This paper compares the performance of modern deep learning models such as YOLOv8, Faster R-CNN, and EfficientDet in object detection in video data. Experiments on the COCO and MOT17 open datasets showed that the YOLOv8 model has the highest speed (145 FPS) and sufficient accuracy (mAP@0.5 = 53.9%), which is superior in real-time applications. At the same time, EfficientDet-D7 (mAP@0.5 = 55.1%) performed best in cases where high accuracy was required.
References
Jocher, G. et al. (2023). Ultralytics YOLOv8. https://github.com/ultralytics/ultralytics
Tan, M. et al. (2020). EfficientDet: Scalable and Efficient Object Detection. CVPR.
Ren, S. et al. (2017). Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks. IEEE TPAMI.
Bewley, A. et al. (2016). Simple Online and Realtime Tracking. ICIP.
Redmon, J. et al. (2016–2023). YOLO seriyasi (v1–v8).
Lin, T.-Y. et al. (2017). Microsoft COCO: Common Objects in Context. ECCV.
Zulunov R., Ergashev O., Akhmadjonov I. Methodology for building license plate recognition systems. Потомки Аль-Фаргани, 2024, 1(1), с. 173-179
Zulunov R., Ergashev O., Akhmadjonov I. The methods of automatic license plate recognition. Потомки Аль-Фаргани, 2024, 1(1), c. 135-141
R.Zulunov, B.Soliyev, A.Kayumov, M.Asraev, Kh.Musayev, D.Abdurasulova. Detecting mobile objects with ai using edge detection and background subtraction techniques. E3S Web of Conferences, 508, 03004 (2024).
R.Zulunov, U.Akhundjanov, B.Soliyev, A.Kayumov, M.Asraev, Kh.Musayev. Building and predicting a neural network in PYTHON. E3S Web of Conferences, 508, 04005 (2024).
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2025 Зулунов Равшанбек Маматович, Асадбек Мухторов

This work is licensed under a Creative Commons Attribution 4.0 International License.
