AI-Enhanced Night Vision Technology Empowers Autonomous Vehicles to Navigate Darkness
-
Science and Technology Daily, Beijing, July 27 (Reporter Zhang Jiaxin) — Thermal images captured by night vision cameras are typically monochromatic, grainy, and somewhat blurry. According to a report published in Nature on the 26th, researchers from Purdue University and Los Alamos National Laboratory have developed a Heat-Assisted Detection and Ranging (HADAR) system. By training artificial intelligence (AI) to determine the temperature, energy signature, and physical texture of each pixel in thermal images, the system produces images nearly as clear as those taken by conventional cameras in daylight, helping autonomous vehicles operate more safely in all outdoor conditions.
To train the AI, researchers used sophisticated thermal imaging cameras and sensors to capture outdoor data at night. These sensors can display energy emissions across the entire electromagnetic spectrum. They also created computer simulations of outdoor environments for additional AI training.
The researchers stated that HADAR has learned to detect objects and estimate their distances with an accuracy ten times higher than relying solely on traditional night vision technology. It utilizes invisible infrared radiation to reconstruct nighttime scenes with clarity comparable to daylight.
However, HADAR still faces practical challenges, such as real-time data acquisition, dynamic blurring, and cost issues. Bulky and expensive cameras and imaging equipment need to be manufactured in smaller sizes and at lower costs. Additionally, the process of collecting and processing data currently takes about a minute, whereas it ideally should be completed within milliseconds to ensure the system's usability in moving autonomous vehicles.