|
Prof. Thrun mentioned that localization under bad weather conditions, like heavy snow, is an open problem at the moment. I'd like to ask Prof. Thrun's opinion about the usage of thermal cameras to help solving this issue: the paved road temperature should always be at list a little higher than the surroundings, and if there is enough traffic on the road, the trails of the previous vehicles should be distinctly warmer, and so easily spotted on the infrared wavelength, both under (heavy?) rain and snow. Maybe these info may be cross-referenced with the landmarks reference system mentioned in the previous office hours discussion (also to be on the safe side, in case the previous car went offroad!). I also have another question about the landmarks sensor system: would it work better using laser scans or camera images? In both case I suppose one could make good use of a database with each record (measurament) indexed by its gps position, but what would be pros and cons in either case? I guess 3D maps build upon range finder measuraments may be affected by piles of snow, while images would certainly need some filtering and probably neural networks to be matched across different weather condition. Also I guess Google maps could be a useful test database for the image solution, while I doubt there is such a vast, worldwide DB for uniform range findings. Thanks! |
|
An IR camera is going to respond to the temperature of the snow, which is fairly uniform. The buried road may be a bit warmer than the grass, but not likely to have an effect through thick snow; if there is a difference, the camera is unlikely to be sensitive enough to detect it. How do aircraft handle bad weather? Radar and wireless beacons. Land vehicles may need a similar trick. If there are tracks of other vehicles, follow the tracks. |
