Efficient Fusion of LiDAR and Camera Data for 3D Object Detection of Intelligent Vehicles

Yingjuan Tang; Hongwen He

doi:10.1109/YAC66630.2025.11150113

Efficient Fusion of LiDAR and Camera Data for 3D Object Detection of Intelligent Vehicles

Yingjuan Tang^*
, Hongwen He

^*Corresponding author for this work

Beijing Institute of Technology

Research output: Contribution to journal › Conference article › peer-review

Abstract

Accurate 3D object detection is vital for autonomous driving, requiring efficient multimodal fusion of LiDAR and camera data. Existing fusion methods often struggle with the tradeoff between accuracy and computational efficiency. We propose UIB-FuseNet, which integrates a universal inverted bottleneck (UIB) fusion module to accelerate fusion while maintaining high detection performance. On the nuScenes dataset, UIB-FuseNet outperforms state-of-the-art methods, improving mAP by 2.78%, NDS by 2.03%, and increasing inference speed by 12.9%, demonstrating its effectiveness for real-time applications in autonomous driving.

Original language	English
Pages (from-to)	586-591
Number of pages	6
Journal	Youth Academic Annual Conference of Chinese Association of Automation, YAC
Issue number	2025
DOIs	https://doi.org/10.1109/YAC66630.2025.11150113
Publication status	Published - 2025
Externally published	Yes
Event	40th Youth Academic Annual Conference of Chinese Association of Automation, YAC 2025 - Zhengzhou, China Duration: 17 May 2025 → 19 May 2025

Keywords

3D object detection
LiDAR and camera data
Multimodal fusion
Universal inverted bottleneck

Access to Document

10.1109/YAC66630.2025.11150113

Cite this

@article{dfe3ab0dfef546f980069df37d6fbe56,

title = "Efficient Fusion of LiDAR and Camera Data for 3D Object Detection of Intelligent Vehicles",

abstract = "Accurate 3D object detection is vital for autonomous driving, requiring efficient multimodal fusion of LiDAR and camera data. Existing fusion methods often struggle with the tradeoff between accuracy and computational efficiency. We propose UIB-FuseNet, which integrates a universal inverted bottleneck (UIB) fusion module to accelerate fusion while maintaining high detection performance. On the nuScenes dataset, UIB-FuseNet outperforms state-of-the-art methods, improving mAP by 2.78\%, NDS by 2.03\%, and increasing inference speed by 12.9\%, demonstrating its effectiveness for real-time applications in autonomous driving.",

keywords = "3D object detection, LiDAR and camera data, Multimodal fusion, Universal inverted bottleneck",

author = "Yingjuan Tang and Hongwen He",

note = "Publisher Copyright: {\textcopyright} 2025 IEEE.; 40th Youth Academic Annual Conference of Chinese Association of Automation, YAC 2025 ; Conference date: 17-05-2025 Through 19-05-2025",

year = "2025",

doi = "10.1109/YAC66630.2025.11150113",

language = "English",

pages = "586--591",

journal = "Youth Academic Annual Conference of Chinese Association of Automation, YAC",

issn = "2837-8598",

number = "2025",

}

TY - JOUR

T1 - Efficient Fusion of LiDAR and Camera Data for 3D Object Detection of Intelligent Vehicles

AU - Tang, Yingjuan

AU - He, Hongwen

PY - 2025

Y1 - 2025

N2 - Accurate 3D object detection is vital for autonomous driving, requiring efficient multimodal fusion of LiDAR and camera data. Existing fusion methods often struggle with the tradeoff between accuracy and computational efficiency. We propose UIB-FuseNet, which integrates a universal inverted bottleneck (UIB) fusion module to accelerate fusion while maintaining high detection performance. On the nuScenes dataset, UIB-FuseNet outperforms state-of-the-art methods, improving mAP by 2.78%, NDS by 2.03%, and increasing inference speed by 12.9%, demonstrating its effectiveness for real-time applications in autonomous driving.

AB - Accurate 3D object detection is vital for autonomous driving, requiring efficient multimodal fusion of LiDAR and camera data. Existing fusion methods often struggle with the tradeoff between accuracy and computational efficiency. We propose UIB-FuseNet, which integrates a universal inverted bottleneck (UIB) fusion module to accelerate fusion while maintaining high detection performance. On the nuScenes dataset, UIB-FuseNet outperforms state-of-the-art methods, improving mAP by 2.78%, NDS by 2.03%, and increasing inference speed by 12.9%, demonstrating its effectiveness for real-time applications in autonomous driving.

KW - 3D object detection

KW - LiDAR and camera data

KW - Multimodal fusion

KW - Universal inverted bottleneck

UR - https://www.scopus.com/pages/publications/105030940152

U2 - 10.1109/YAC66630.2025.11150113

DO - 10.1109/YAC66630.2025.11150113

M3 - Conference article

AN - SCOPUS:105030940152

SN - 2837-8598

SP - 586

EP - 591

JO - Youth Academic Annual Conference of Chinese Association of Automation, YAC

JF - Youth Academic Annual Conference of Chinese Association of Automation, YAC

IS - 2025

T2 - 40th Youth Academic Annual Conference of Chinese Association of Automation, YAC 2025

Y2 - 17 May 2025 through 19 May 2025

ER -

Efficient Fusion of LiDAR and Camera Data for 3D Object Detection of Intelligent Vehicles

Abstract

Keywords

Access to Document

Other files and links

Fingerprint

Cite this