Adaptive Dynamic Resolution Detection for Real-Time Video Streams
ITM Web of Conferences · 2026 · 89 · 01009
DOI: 10.1051/itmconf/20268901009 ↗Kurz gefasst
Fachartikel in der Zeitschrift ITM Web of Conferences, online erschienen am 14.08.2026. Verfasst von Yiluan Huang. Die Arbeit nennt 10 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
Real-time video object detection on edge devices faces the challenge of limited computing resources, and higher resolution leads to greater computational costs. This paper improves the static strategy in the DynaPP method, which fixes the key frame interval and processes non-key frames with a fixed low resolution. A self-adaptive dynamic resolution detection framework is proposed. This framework consists of two lightweight modules: an adaptive key frame selector based on content complexity (AKS) and a lightweight resolution predictor (LRP). Experimental results on the VisDrone dataset show that the AKS module can increase the processing speed by 5.5% (from 82.02 FPS to 86.53 FPS), while maintaining the detection performance. The LRP module achieves dynamic resolution selection with almost no loss in speed. However, on high-performance GPUs, due to the Canvas packaging overhead and sufficient computing resources, the benefits of resolution reduction are limited. Through in-depth analysis of the VisDrone dataset, it is found that target density and scale have a significant impact on the acceleration effect, which points out the direction for subsequent optimization.
Zitieren
APA
Huang, Y. (2026). Adaptive Dynamic Resolution Detection for Real-Time Video Streams. ITM Web of Conferences, 89, 01009. https://doi.org/10.1051/itmconf/20268901009
Vancouver
Huang Y. Adaptive Dynamic Resolution Detection for Real-Time Video Streams. ITM Web of Conferences. 2026;89:01009. doi:10.1051/itmconf/20268901009
BibTeX
@article{Huang2026,
author = {Huang, Yiluan},
title = {Adaptive Dynamic Resolution Detection for Real-Time Video Streams},
journal = {ITM Web of Conferences},
year = {2026},
volume = {89},
pages = {01009},
doi = {10.1051/itmconf/20268901009}
}
EUVIMED