Skip to main navigation Skip to search Skip to main content

Innovative approaches in image-based 3D object detection for autonomous driving: A comprehensive review

  • Malik Haris
  • , Yaoguo Zhang
  • , Guoqiang Zhang*
  • , Muhammad Shahid Mastoi
  • , Mannan Hassan
  • , Asif Raza
  • , Zhengqing Li
  • *Corresponding author for this work

Research output: Contribution to journalReview articlepeer-review

3 Scopus citations

Abstract

Two-dimensional object detection (2DOD) is a renowned discipline, but it fundamentally lacks the essential depth and dimensional information required for safe autonomous driving. To overcome this limitation, multiple researchers have proposed image-based three-dimensional object detection (3DOD) approaches. This paper presents a comprehensive review of recent progress in this cost-effective field, with a particular focus on the underlying methodologies. We present an innovative, granular classification framework for these image-based approaches and categorize the literature into Image-based 3D object detection (IM-3DOD) and Fusion of point cloud-based and image-based 3D object detection (F-PCIM-3DOD) . This manuscript provides a summary of important datasets, including those from the Karlsruhe Institute of Technology and Toyota Technological Institute (KITTI), nuScenes , and Waymo Open datasets, and outlines standard evaluation metrics. Our analysis highlights that while effective F-PCIM-3DOD can significantly improve 3DOD accuracy, with top-performing models now exceeding 87% mean Average Precision (mAP) on the KITTI "Car" benchmark, performance remains severely limited for small or distant objects like pedestrians and cyclists. We conclude by discussing several intriguing areas for future study, including singular pooling, semi-supervised training, and real-time fusion strategies, which are crucial for navigating the inherent trade-off between accuracy and computational speed in autonomous systems. All data in this research paper is available at: https://github.com/malikharispk/Innovative-Approaches-3D-Object-Detection .

Original languageEnglish
Article number106016
JournalDigital Signal Processing: A Review Journal
Volume175
DOIs
StatePublished - 15 May 2026
Externally publishedYes

Bibliographical note

Publisher Copyright:
© 2026 Elsevier Inc.

Keywords

  • 3D object detection (3DOD)
  • Image-based detection
  • Multimodal fusion
  • Point cloud
  • Spectral pooling
  • Transformer networks

ASJC Scopus subject areas

  • Signal Processing
  • Computer Vision and Pattern Recognition
  • Statistics, Probability and Uncertainty
  • Computational Theory and Mathematics
  • Artificial Intelligence
  • Applied Mathematics
  • Electrical and Electronic Engineering

Fingerprint

Dive into the research topics of 'Innovative approaches in image-based 3D object detection for autonomous driving: A comprehensive review'. Together they form a unique fingerprint.

Cite this