Switch off the rain please, I am driving….
Monday, July 16th, 2012 by Roberto Saracco
Predicting and avoiding illumination of a single drop. In this experiment, the drop is first imaged with 5ms camera exposure time (shown above magenta dashed line). The system latency is shown above the cyan dashed line. Images shown are composites of the 14 frames needed for the drop to traverse the field of view (FOV). Since the experiments are repeatable, the ground-truth image on the left shows drops illuminated throughout the entire FOV (a). The drop is falling with near constant velocity at 16 pixels per frame, so prediction is straightforward (shown as yellow boxes) (b). The light would be projected during the dark frames. (Credit: CMU)
Well, researchers imagination never stops to amaze me. I run onto a prototype by CMU researchers, working in the CMU Robotic Institute, that have created a system to cancel the rain, or snowflakes, from your vision as you drive your car! Just flick a switch and, like magic, the rain disappear from your windshield.
The idea is based on the fact that when we drive under a squall, or a snow storm, the beams light is reflected by the rain drops and snow flakes. This reflection makes them visible to our eyes and creates a sort of halo, a persistency, on our retina that confuse the image.
The researchers placed a camera on the windshield and have connected it to a computer that “sees” the raindrops and snowflakes. Hence it continuously redirect the car beams to avoid illuminating the drops or flakes. They have shown that their system is able to detect a raindrop, predict its movement and redirect the beam in just 13ms.
Our eyes are not able to see the blinking of the lights as it gets redirected and get a clear image of the road since it is no longer obstructed by the illuminated raindrops.
The effect is amazing, and you should take a look at the video showing what is like driving in a rainstorm without or with this system:
http://www.cs.cmu.edu/%7EILIM/projects/IL/smartHeadlight/media/vids/inVehicleSnow_hr.mp4
Tags: cars, image recognition, sensors