Skip to main navigation Skip to search Skip to main content

A Pseudo-LiDAR and Image Fusion Method for Autonomous Driving

  • Yalong Ma
  • , Xin Gao
  • , Ziying Yao
  • , Xinkai Wu*
  • *Corresponding author for this work
  • Beihang University

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Autonomous vehicles typically integrate LiDAR sensors for precise but sparse distance measurements and camera sensors for rich visual data, albeit without 3D object location. This paper introduces a fusion algorithm designed to incorporate pseudo-LiDAR information generated from RGB images with LiDAR data and image edges. Firstly, the approach extracts depth estimates from RGB images to produce dense yet somewhat noisy pseudo-LiDAR data. Subsequently, it establishes connections between pseudo-LiDAR points and LiDAR points, incorporating image edges for denser and more accurate fusion data. The resultant fused data can then be utilized as input for other 3D object detection networks, yielding enhanced performance in multi-sensor fusion perception. Experimental results on the KITTI benchmark corroborate the effectiveness of the proposed method.

Original languageEnglish
Title of host publicationCICTP 2024
Subtitle of host publicationResilient, Intelligent, Connected, and Lowcarbon Multimodal Transportation - Proceedings of the 24th COTA International Conference of Transportation Professionals
EditorsJianming Ma, Qin Luo, Lijun Sun, Baicheng Li, Jingjing Chen, Guohui Zhang
PublisherAmerican Society of Civil Engineers (ASCE)
Pages45-54
Number of pages10
ISBN (Electronic)9780784485484
DOIs
StatePublished - 2024
Event24th COTA International Conference of Transportation Professionals: Resilient, Intelligent, Connected, and Lowcarbon Multimodal Transportation, CICTP 2024 - Shenzhen, China
Duration: 23 Jul 202426 Jul 2024

Publication series

NameCICTP 2024: Resilient, Intelligent, Connected, and Lowcarbon Multimodal Transportation - Proceedings of the 24th COTA International Conference of Transportation Professionals

Conference

Conference24th COTA International Conference of Transportation Professionals: Resilient, Intelligent, Connected, and Lowcarbon Multimodal Transportation, CICTP 2024
Country/TerritoryChina
CityShenzhen
Period23/07/2426/07/24

Fingerprint

Dive into the research topics of 'A Pseudo-LiDAR and Image Fusion Method for Autonomous Driving'. Together they form a unique fingerprint.

Cite this