Driving in a snowstorm at night is incredibly difficult and stressful. Over 88,000 crashes caused by snow or sleet occur at night resulting in hundreds of fatalities (NHTSA Safety Stats). Snowflakes are illuminated brightly and distract the driver from observing the entire road (similar to rain). Researchers in computer vision have proposed methods for removing snow from videos. Processed videos can be displayed for the driver, but current implementations are not intuitive and, at times, distracting for the driver.
We address this problem with a solution similar to that for anti-glare, i.e., reacting to detected bright objects. The main difference, however, is that the density, size, and speed of snowflakes requires high-resolution, low-latency illumination to be effective. Therefore, we exploit the high spatiotemporal beam control capabilities of our prototype to distribute light between falling snowflakes to reduce backscatter directly in the driver's visual field. [direct link to illustration above]
Distracting snow streaks while driving with high and low beams (download video).
Our Approach and Results
Concept: Snowflakes are detected and light is streamed in between them to reduce the amount of light reflected back to the driver.
Seeing Through Snow: Backscatter from standard headlights appear as long, flickering streaks due to the human eye's long integration time. Streaming light between snowflakes dramatically reduces their visibility. Note: Artificial snowflakes used in these videos. [direct video link]
Performance Evaluation: Our current prototype can achieve 98% visibility improvement with less than 10% light loss in light snow and 40% light loss in heavy snow (stationary vehicle). Performance of our our previous prototype (vehicle moving 30 kph) is also shown.
