This repository provides a comprehensive survey of 3D Occupancy Perception for Autonomous Driving, focusing on information fusion techniques. It targets researchers and engineers in autonomous driving and computer vision, offering a structured overview of the field, including methodologies, datasets, and applications.

How It Works

The survey systematically categorizes 3D occupancy perception methods into LiDAR-centric, Vision-centric, Radar-centric, and Multi-Modal approaches. It details network pipelines, fusion techniques, and training strategies, providing in-depth analyses and performance comparisons. The repository also curates relevant datasets and discusses various occupancy-based applications like segmentation, detection, tracking, and scene generation.

Quick Start & Requirements

This repository is a survey and does not contain executable code for a specific model. It links to numerous research papers, many of which provide code repositories for their respective implementations. Requirements vary per linked project.

Highlighted Details

• Systematically surveys the latest research on 3D occupancy perception in autonomous driving.
• Provides a taxonomy of perception methods, elaborating on core issues like network pipelines, fusion, and training.
• Includes detailed performance evaluations and comparisons, alongside discussions on current limitations and future directions.
• Accepted by Information Fusion (Impact Factor: 14.7), with over 192 literature references.

Maintenance & Community

This is an active repository, regularly updated with new research. Contributions and suggestions are welcomed via pull requests or direct contact. The primary contact is Professor Lap-Pui Chau.

Licensing & Compatibility

The repository itself is not licensed as it is a collection of survey information and links. Individual linked projects will have their own licenses.

Limitations & Caveats

As a survey, this repository does not offer a single, unified implementation. Users must refer to individual linked papers for specific code, dependencies, and usage instructions. The rapid pace of research means the survey is a snapshot in time.