The hidden moments of one of the most popular American television programs : Shark Tank

 How artificial intelligence improves the sharpness of fuzzy thermal vision photos

 The blurry, ghostly shapes often seen in thermal imaging could be a thing of the past. By combining artificial intelligence and thermal imaging, scientists can create sharp, detailed images even in the dark. The technology could one day help improve the nighttime navigation capabilities of autonomous vehicles.

Thermal imaging, commonly used in night vision systems, works by detecting heat sources. Infrared images appear blurry due to a phenomenon called ghosting. The heat of an object overwhelms every detail of its texture, just as turning on a light makes it difficult to decipher the engraving on a lightbulb.

 

  In, theoretical physicist, Fanglin Bao of Purdue University in West Lafayette, Indiana, and his colleagues used a thermal imaging camera that can distinguish between different wavelengths of infrared light. The researchers paired this camera with a computer program that uses AI to decode information from the device, revealing the temperature, texture, and material type of objects in an image. This technique drew bright, detailed images from dark night scenes, the team reports in Nature on July 26.

 

"There are no restrictions on adverse weather conditions or night-time scenarios," says electrical engineer Muhammad Ali Farooq of Galway University in Ireland, who was not involved in the study."Even in poor lighting conditions, very good and sharp data can be obtained."

The technique can also measure distances with about the same precision as current camera-based methods. This means it could be used in autonomous vehicles that need to know when to brake to avoid an accident (SN: 10/12/18).

Today's autonomous vehicles often measure distances by bouncing signals off objects, much like how sonar works. Many autonomous cars sending signals could be confused with one another. Because the new tech doesn't need to send a signal, it could be safer to expand in a world with more autonomous vehicles, the researchers say.

Nevertheless, the technology will not find its way onto busy roads in the foreseeable future. The camera is heavy, measuring about 2 feet on each side, and expensive, costing more than $1 million, Bao says. Each image takes about a second to capture, which is too slow for an autonomous vehicle that needs to react to situations in real-time.

 

Still, Bao hopes to see which versions of this technology could be used for autonomous vehicles or robots in the future."Humans have an evolutionary preference for light," he says. "But it turns out that AI can overcome this age-old dichotomy between day and night."