Seeing Color In Darkness: Using AI To Add Color To Night Vision
Night vision technology that has been used for decades, but it's always had the limitation of not being able to provide color. A new report seeks to bring color to night vision with the help of deep learning.
Created on April 8|Last edited on April 9
Comment
Infrared night vision cannot produce color the way humans see it, as it relies on infrared light which humans cannot percieve. To work with this limitation, the infrared light must be interpreted by a program and displayed to a screen, usually a grayscale image sometimes tinted green, so that humans may see what the camera sees.
A new report published by Andrew W. Browne et al. seeks to add a step to that interpretation computation and bring color to infrared night vision through deep learning.
How does it work?
The objective for the teaching process is to be able to add color to infrared pictures by training a deep learning model.
First they needed to collect the data to train on, a dataset comprised of many pictures, including visible light spectrum photos and infrared pictures not able to be seen by humans. They took numerous pictures at very precise wavelengths and then compiled these pictures into datasets for the learning process.
Three wavelengths of each light spectrum were selected and act as analogues of eachother, giving us a sort of "RGB" of the infrared light spectrum. Using this infrared "RGB", the model can be trained to convert the values of those wavelengths to a color image through.
Going forward
This is just the first step for full-color infrared technology. Further work is needed to refine the accuracy, but as a start, the results are amazing. The technology is not coming to infrared goggles any time soon, but it's something to look forward to.
Find out more
Add a comment
Tags: ML News
Iterate on AI agents and models faster. Try Weights & Biases today.