Toby Murray-2 wrote > > Well if you really want harsh lighting, you might look at the video I > just uploaded the other day from biking across Kansas. > Wow, those really are difficult lighting conditions! Even as a human one has trouble reading the speed limit signs in those videos.
Apart from the Sun, those videos seem like some ideal test cases for this system. High quality, and the type of rural road where this system could be most useful. Very few POIs and then the odd road sign every couple of minutes. So automatically condensing those 18 minutes into a few relevant frames that have interesting content could really be a time saver. Toby Murray-2 wrote > > It is full > motion video too, not timelapse. > I think under these conditions, the benefit of full motion video vs timelapse can be particularly striking. Most of the time the road signs are not readable due to the sun and only at very specific angles can you discern the writing on the signs. With time laps video, chances are that one would miss the short time frame where one can comparatively easily read the signs. The full motion pictures might also allow for some other tricks to be played. With the help of optical flow one can probably determine the depth / distance from the camera and thus the size of objects. As street signs have a well defined size, one could potentially use that as extra heuristics to better identify road signs under these difficult conditions. Once one has locked onto a sign and can track it through the video one can either choose the best frames, or use classifiers that average across the multiple frames in which the sign is viewable, hopefully increasing the accuracy of the system. Depending on how Nikhil gets along, these kind of fancy computer vision tricks are probably beyond the scope of the project though. But it would be cool if one could try some of this out... Kai -- View this message in context: http://gis.19327.n5.nabble.com/GSoC-Video-based-speed-limit-and-roadsign-detector-and-Request-for-videos-of-videomapping-tp5709573p5714254.html Sent from the Developer Discussion mailing list archive at Nabble.com. _______________________________________________ dev mailing list dev@openstreetmap.org http://lists.openstreetmap.org/listinfo/dev
