EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

BEV-Nexus: BEV Perception Algorithm Based on Depth Perception Enhancement and Dynamic Adaptive Fusion

Xiaona Song, Haozhe Zhang, Zhengyi Huang, Jianlin Zhao, Lijun Wang

Sensors · 25.07.2026 · 26 (15) · 4720

DOI: 10.3390/s26154720 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Sensors, online erschienen am 25.07.2026. Verfasst von Xiaona Song und 4 weiteren Autor:innen. Beteiligte Einrichtung: School of Mechanical Engineering, North China University of Water Resources and Electric Power, Zhengzhou 450045, China. Die Arbeit nennt 50 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, the Scientific and Technological Project of Henan Province, Key Research and Development Projects in Henan Province.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This paper proposes an improved multimodal fusion framework for 3D object detection, termed BEV-Nexus, which aims to address the issues of inaccurate depth estimation and inefficient fusion paradigms in existing image-point cloud fusion methods. We introduce a Point-Cloud-Guided Depth Prediction Network (PCGD-Net), which enhances the image branch’s depth prediction capability by embedding point cloud spatial prior, ground-truth loss constraint, and projected point cloud depth filling. Additionally, we design a Dynamic Self-adaptive Feature Fusion Module (DSF-Module), which computes multimodal feature similarity using window attention and performs weighted fusion based on self-adaptive weights, resolving alignment deviations in BEV features. Finally, we propose a Dilated Attention Enhancement Block (DAEB), which expands the receptive field through dilated convolution and integrates parameter-free attention mechanism (SimAM) for feature enhancement, ensuring efficiency while improving overall feature representation. Experimental results on nuScenes validation set show that BEV-Nexus outperforms it baseline (BEVFusion) by 1.8% mAP and 1.5% NDS. On the test set, BEV-Nexus improves mAP and NDS by 1.6% and 1.4%, respectively. Furthermore, the detection FPS remains nearly unchanged, demonstrating significant lightweight advantages.

Schlagwörter

Stichwörter der Autor:innen

3D object detectionBEV (bird’s eye view)LiDARautonomous drivingmultimodal fusion

Autor:innen und Institutionen

  • School of Mechanical Engineering, North China University of Water Resources and Electric Power, Zhengzhou 450045, China.

Förderung

  • National Natural Science Foundation of China
  • the Scientific and Technological Project of Henan Province
  • Key Research and Development Projects in Henan Province
  • Water Conservancy Equipment and Intelligent Operation & Maintenance Engineering Technology Research Centre in Henan Province
  • Graduate Education Reform Project of Henan Province
  • Henan International Joint Laboratory of Thermo-fluid Electro-chemical System for New Energy Vehicle
  • Scientific and Technological Project of Henan Province

Zitieren

APA

Song, X., Zhang, H., Huang, Z., Zhao, J., & Wang, L. (2026). BEV-Nexus: BEV Perception Algorithm Based on Depth Perception Enhancement and Dynamic Adaptive Fusion. Sensors, 26(15), 4720. https://doi.org/10.3390/s26154720

Vancouver

Song X, Zhang H, Huang Z, Zhao J, Wang L. BEV-Nexus: BEV Perception Algorithm Based on Depth Perception Enhancement and Dynamic Adaptive Fusion. Sensors. 2026;26(15):4720. doi:10.3390/s26154720

BibTeX
@article{Song2026,
  author = {Song, Xiaona and Zhang, Haozhe and Huang, Zhengyi and Zhao, Jianlin and Wang, Lijun},
  title = {BEV-Nexus: BEV Perception Algorithm Based on Depth Perception Enhancement and Dynamic Adaptive Fusion},
  journal = {Sensors},
  year = {2026},
  volume = {26},
  number = {15},
  pages = {4720},
  doi = {10.3390/s26154720}
}
Kostenlos starten →