EUVIMEDEUVIMED entdecken
Crossref

Analysis of the architectural evolution and hardware-oriented improvements of UAV systems for real-time object detection

А. А. Піпіч, Б. В. Булах

Scientific Bulletin of UNFU · 25.06.2026 · 36 (3)

DOI: 10.36930/40360313 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Scientific Bulletin of UNFU, online erschienen am 25.06.2026. Verfasst von А. А. Піпіч und 1 weiteren Autor:innen. Die Arbeit nennt 14 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This survey examines the real-time object detection by UAVs between 2024 and 2026 years and explores its critical balance between accuracy, inference speed, and power consumption under strict hardware limitations. These hardware limitations are defined by typical conditions that are expected when mentioned object detection modules are mounted on UAVs. Survey also explores how modern architectures of real-time object detection models evolved from classic CNN-centric designs (YOLOv11) to attention-centric paradigms (YOLOv12) and latest solutions with native NMS-free inference (YOLO26). It also explores how these architectures can suffer from significant engineering challenges, such as "heavy" architectural blocks that risk overflowing memory bandwidth on entry-level edge platforms. The survey examines the most critical research gaps of an industry of real-time object detection by UAVs. Among the existing research gaps in this area two major ones were identified. The first one is related to an observation: approximately 90 % of the newest models (YOLO12, YOLO26, TSS-YOLO, other NMS-Free inference optimized architectures) is missing the verified energy efficiency or energy consumption (by the hardware required for mounting a specific model) data. The second one is the "Dataset Dependency" problem: general benchmarks like MS-COCO are not directly comparable to specialized aerial results. The survey reviews specialized solutions like RTUAV-YOLO and LiteCOD that utilize architectural ingenuity (such as dynamic weight generation and efficient spatial attention) to bridge the massive 8 x performance gap (TOPS) found across different hardware tiers. To address mentioned discrepancies, the survey evaluates advanced optimization strategies, including mixed-precision Pareto optimization (PoMQ-ViT), hardware-aware Neural Architecture Search (NAS), and structural branch removal (TSS-YOLO), which is found to be more effective than standard quantization alone. The introduction of Unified Complexity Coefficient is proposed in this article to resolve an issue of performance normalization and allow to compare results of general benchmarks and specialized solutions. The survey is also predicting a potential industry shift toward adaptive, "compression-first" systems, which may finally ensure true real-time independency and autonomy for the next generation of autonomous aerial platforms.

Zitieren

APA

Піпіч, А. А., & Булах, Б. В. (2026). Analysis of the architectural evolution and hardware-oriented improvements of UAV systems for real-time object detection. Scientific Bulletin of UNFU, 36(3). https://doi.org/10.36930/40360313

Vancouver

Піпіч АА, Булах БВ. Analysis of the architectural evolution and hardware-oriented improvements of UAV systems for real-time object detection. Scientific Bulletin of UNFU. 2026;36(3). doi:10.36930/40360313

BibTeX
@article{2026,
  author = {Піпіч, А. А. and Булах, Б. В.},
  title = {Analysis of the architectural evolution and hardware-oriented improvements of UAV systems for real-time object detection},
  journal = {Scientific Bulletin of UNFU},
  year = {2026},
  volume = {36},
  number = {3},
  doi = {10.36930/40360313}
}
Kostenlos starten →