Related Experiment Video
Updated: Dec 8, 2025
13:38
Laser-Induced Fluorescence Emission L.I.F.E. as Novel Non-Invasive Tool for In-Situ Measurements of Biomarkers in Cryospheric Habitats
Published on: October 26, 2019
8.3K
Color Image Generation from Range and Reflection Data of LiDAR
Hyun-Koo Kim1, Kook-Yeol Yoo1, Ho-Youl Jung1
1Department of Information and Communication Engineering, Yeungnam University, Gyeongsan 38544, Korea.
Sensors (Basel, Switzerland)
|September 24, 2020
Summary
This study demonstrates generating color images from LiDAR range data by fusing it with reflection data using deep learning. The proposed "last fusion" method significantly improves image quality and enables real-time applications.
Area of Science:
- Computer Vision
- Robotics
- 3D Sensing
Background:
- 3D Light Detection and Ranging (LiDAR) typically provides range and reflection data.
- Generating realistic color images from LiDAR data is an active research area.
- Existing methods primarily use LiDAR reflection data for image generation.
Purpose of the Study:
- To generate color images from LiDAR range data.
- To fuse LiDAR reflection and range data using deep learning for enhanced image generation.
- To evaluate different data fusion strategies (early, mid, last) for this task.
Main Methods:
- Proposed deep learning networks utilizing an encoder-decoder structured fully convolutional network (ED-FCN).
- Implemented early, mid, and last fusion techniques to combine reflection and range data.
- Trained and verified models using the KITTI dataset.
Main Results:
- The 'last fusion' method outperformed reflection-only and range-only methods.
- Achieved improvements of 0.53 dB (grayscale PSNR), 0.49 dB (color PSNR), and 0.02 (SSIM) compared to the baseline.
- Demonstrated real-time applicability with an average processing time of 13.56 ms per frame.
Conclusions:
- Fusing LiDAR range and reflection data significantly enhances color image generation quality.
- The proposed 'last fusion' deep learning approach is effective and suitable for real-time applications.
- This methodology offers a powerful tool for multi-source heterogeneous data fusion.
