Detecting Objects in the Dark

What is it about?

We all love to take photographs in the night and use them for downstream applications such as driving, pedestrian identification and robot navigation. But traditionally such applications were possible only in good lighting conditions such as daytime. This is because Nighttime photographs are susceptible to poor visibility and high noise. In this work we thus propose a technique to enable above applications even in extremely dark conditions in a time and resource efficient manner.

Featured Image

Photo by Ryan Hutton on Unsplash

Photo by Ryan Hutton on Unsplash

Why is it important?

Night time applications are not possible with regular RGB cameras and requires specialized hardware like NIR/thermal sensors, LiDAR's, etc. This increases the bill of material and generally these devices are very bulky and power hungry. Our technique uses artificial intelligence to solve much of the nighttime problems with regular RGB cameras for time and memory efficiency, critical for real world applications. Finally since RGB cameras are readily available today even for casual photographers and lay man in the form of smartphone cameras, we believe our work empowers a much larger audience and no longer the field will be restricted to people who can afford costly equipments and makes the technology more accessible wherever it is not possible to assemble specialized but bulky apparatus.