|
Active Sensing/Illumination Systems for Cars
|
Imagine you are driving at night and it is raining hard. The rain, the splashes, the drops on the
windshield and the shiny wet road all cause significant degradation of scene visibility, making
driving an unpleasant experience at best and a dangerous endeavor at worst.
Instead of removing rain from images/videos and showing the results on a display, can we somehow illuminate the scene, such that the visibility of rain is reduced for a human observer?
|
|
How it Works
|
Overall Approach: The headlight consists of a camera pair and a DLP projector. The camera pair is used to perform 3D reconstruction of detected drop positions. Future locations of drops are predicted by tracking detected drops then fitting a second order model to match to a nearest-neighbor cluster in a look-up table. The entire process from capture to reaction takes about 70 ms. |
|
|
Detecting, predicting and tracking drops: Intermediate steps for the drop detection algorithm. We used four drop emitters, each emitting at 2 Hz. |
| |
| System at Work: These images are taken at two different times, ?rst with the projector displaying white, then with the projector detecting, tracking, and blacking-out the drops. (To show long streaks, both images are composites of multiple images, and the speckled appearance of the eight streaks is due to the DLP projector's dithering). For (b), the tops of the streaks are white, as the system has about 70ms latency between the camera and projector. But once detected and tracked, and the projector displays black, the drops are much dimmer. They are still visible, as the projector causes global illumination effects, which the drops re?ect and refract. |
| |
|
Rain Generator: The basic adaptive projection testbed for generating rain drops indoors.
|
|
Publications
"Light and Water Drops"
Peter Barnum, Doctoral dissertation, Tech. Report, Robotics Institute, Carnegie Mellon University, May, 2011 [pdf]
|
|
|
|
|
