RGBT-1K: An RGB-Thermal Paired dataset

Citation Author(s):
Santosh Kumar
Panda
National Institute of Technology Rourkela
Pankaj Kumar
Sa
National Institute of Technology Rourkela
Submitted by:
Santosh Panda
Last updated:
Tue, 11/05/2024 - 04:48
DOI:
10.21227/dtt8-vh83
Data Format:
License:
0
0 ratings - Please login to submit your rating.

Abstract 

In this dataset, we present a novel RGB-Thermal paired dataset, RGBT-1K, comprising 1,000 image pairs specifically curated to support research in multi-modality image processing. The dataset captures diverse indoor and outdoor scenes under varying lighting conditions, offering a robust benchmark for applications in image enhancement, object detection, and scene analysis. The image acquisition process involved using the FLIR A70 thermal camera and the Sony Handycam HDR-CX405, with the latter positioned atop the thermal camera for precise alignment. The dataset then goes through image registration and transformation steps to better align it pixel by pixel for training in learning-based approaches. This setup ensured synchronized RGB and thermal captures, allowing for accurate multi-modal data analysis. Potential applications of this dataset span a wide range of computer vision tasks, including surveillance, autonomous navigation, and medical imaging, making it a valuable resource for academia and industry.

Instructions: 

The images provided are not registered. Users can register the images using the steps provided in the documentation file. You can copy the code, paste it into MATLAB, and directly run it, giving appropriate paths of images. For study purposes, we have provided a subset of the dataset comprising 50 image pairs ( 100 total images ).

Funding Agency: 
DST INSPIRE, Govt of India
Grant Number: 
IF190566

Documentation

AttachmentSize
File Image registration steps in MATLAB.txt435 bytes