Media release

From: Springer Nature

Technology: Heat assisted detection makes night as clear as day (N&V)

A new technology that incorporates heat detection to enhance computer vision and ranging systems is described in Nature this week. The approach enables improved perception and identification of objects at night and may have applications in autonomous vehicle navigation.

The ability of machines to perceive their surroundings is pertinent to social interactions between humans and robots (including automated vehicles and robot helpers), enabling these machines to make decisions without human intervention. Camera vision relies on illumination, so many machines augment camera vision using sensors that work with non-visible information such as sonar, radar and LiDAR. Heat signals can also be used to assist perception of a scene, but because objects constantly emit and scatter thermal radiation, thermal vision tends to lack material specificity and can result in hazy, indistinct and textureless images.

Zubin Jacob and colleagues developed ‘heat-assisted detection and ranging’ (HADAR) that uses thermal physics and machine learning to improve night-time sensing. HADAR uses techniques that disambiguate cluttered heat signals. The authors demonstrate that HADAR can sense texture and depth through darkness, and that it can perceive physical attributes. HADAR is more accurate than currently available thermal ranging techniques at night, and in daylight it shows an accuracy comparable to conventional illumination-based colour stereovision in the day.

The authors claim that, while HADAR still faces practical challenges (such as data acquisition in real time, motion blur and cost), it may have wide applications including in automated navigation and wildlife monitoring.