Related Experiment Video
Updated: Aug 17, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
604
A Critical Review of Deep Learning-Based Multi-Sensor Fusion Techniques
Benedict Marsh1, Abdul Hamid Sadka1, Hamid Bahai2
1Institute of Digital Futures, Brunel University London, Kingston Ln, Uxbridge UB8 3PH, UK.
Sensors (Basel, Switzerland)
|December 11, 2022
Summary
This review covers deep learning methods for multi-sensor fusion, combining RGB stereo images and LiDAR depth maps to create dense depth predictions. We evaluate current techniques, highlighting their strengths and ideal applications.
Area of Science:
- Computer Vision
- Robotics
- Sensor Fusion
Background:
- Depth map prediction is crucial for autonomous systems.
- Integrating RGB stereo and LiDAR data offers richer environmental perception.
- Traditional methods struggle with dense, accurate depth estimation.
Purpose of the Study:
- To review state-of-the-art multi-sensor fusion techniques for dense depth map prediction.
- To analyze deep learning-based, end-to-end trainable fusion methods.
- To provide a comparative evaluation of current fusion approaches.
Main Methods:
- Coverage of multi-sensor fusion techniques using RGB stereo images and sparse LiDAR depth maps.
- Focus on recent deep learning-based, end-to-end trainable methods.
- Comparative analysis of identified state-of-the-art techniques.
Main Results:
- Identification of leading deep learning-based multi-sensor fusion techniques.
- Detailed analysis of the strengths and limitations of each method.
- Assessment of suitability for various real-world applications.
Conclusions:
- Deep learning significantly advances dense depth map prediction from multi-sensor inputs.
- The choice of fusion technique depends on specific application requirements and performance trade-offs.
- Further research can optimize fusion strategies for enhanced accuracy and efficiency.
