Various activities require a person to see in the dark. Unfortunately, humans have poor night vision compared to the other members of the animal kingdom, so several technologies have been developed to aid the human eye in seeing images clearly in the dark.
Most night vision devices consist of conventional cameras to capture the ambient visible light reflected by objects. Some use radar and lidar, sending radio waves and laser beams to map the surroundings. However, night vision devices face challenges since cameras cannot capture images without reflected light, while radar and lidar can be prone to interference. Another technology, thermal imaging, provides a better approach, although it also faces barriers in image recognition.
Challenges in Thermal Imaging
Thermal imaging refers to reconstructing a scene by capturing the infrared light emitted by objects in the environment. It was initially developed for military operations during the Korean War in the 1950s.
Every object emits infrared energy, which is also called heat signature. The thermal camera, also known as an infrared camera, works by detecting and measuring objects' infrared radiation. This energy is converted into an electronic image showing the surface temperature of the object being seen.
Today, thermal imaging cameras have been extensively used in many other fields, such as transport navigation, healthcare, law enforcement, animal management, and building construction. However, this technology has been underexplored for assisting computer vision.
Since every object releases heat signals, a conventional thermal camera seems to capture a landscape where every item is aglow. As described by electrical and chemical engineer Zubin Jacob from Purdue University, many signals, including noise and clutter, can enter the camera together. Because of this, there is a ghosting effect in the images produced by thermal cameras where the objects appear vague with a low-resonance appearance that is hard to tell apart.
Using AI for Clear Night Vision Images
To address this problem, Jacob and his colleagues use artificial intelligence to develop a neural network program. It works by sorting through the infrared signal from an ordinary thermal camera to separate the characteristic heat signature of an object from the noise and clutter from the environment.
To make this possible, the research team trained the algorithm to recognize the unique emission spectra of every object, like wood, glass, or fabric. As the algorithm identifies the known signatures within a scene, it can characterize what it observes. By assessing the reflection and scattering of environmental noise, the algorithm can generate information about the texture of an object and furnish the image with a higher level of detail.
The processed image not only shows individual objects and their textures, but it can also identify the composition of each material. Jacob and his team are confident that this imaging technique can provide the key data autonomous vehicles need. They also believe it will be an important element of future machine vision technology.
Check out more news and information on Infrared in Science Times.
