YouTube - Edge detector Demo with second stage feature filter
I suck at public speaking when I know its being recorded :D
Printable View
YouTube - Edge detector Demo with second stage feature filter
I suck at public speaking when I know its being recorded :D
Cool :D
When you put the tank in, you said it sees the tree more as noise and the tank as much more interesting than the tree. But, IMO the tree looks better when the tank is behind it :)
It would be cool to run a windy forest (with leaves and branches) through it. For example, if you wanted to identify a tank in a forest ;)
Cool video BTW.
Ill try to get out to a forest to do just that, although it will be my truck instead of a tank, I just need to find a cheap laptop and a webcam that can handle outdoor light levels. Maybe if I get a job with Raytheon :D I suppose I could just take snapshots and feed them through the filter. Or if I wanted to get really fancy I could just take a video, then burn it to DVD then play it through my video capture board, which would show the real time processing. Yeah I think thats what Ill do.
Pretty cool. You neglected to mention the resolution of the feed, however. I'd like to know how a higher resolution feed might effect the filter's judgment on noise vs edges. I'd also like to know how it handles compressed video stored on the disk. There are some video codecs that could bring that processor to a crawl with a decent resolution... let alone reprocessing it three more times.
Being able to pipe any video through your software would "open the flood gates" for testing opportunities.
gg
The feed I'm using is 320x240, processing it 2 times (both the edge detector and feature selector) to determine the filters usage shows that the application would use about 60% of the processor, meaning the rough estimate is the filter itself uses about 10% for 320x240 or 40% at 640x480. The filter isn't currently written to use multiple threads, but it would be trivial to do so.
The capture class I am currently using will capture video from any WDM compliant source, webcams, video capture boards, etc. This makes it possible to feed direct video from say a DVD or a camcorder to the application.
So I should just hang some tree leaves off of my tank?
Very interesting. One thing that I didn't understand however...
First thing that came to mind was applications for the medical field. Dunno, real-time edge detection seems really neat when applied to certain types of tomography scans... I think. The sharpness of your filter was something to behold. But the fact the tree edges where they touch the tank became features of the tank seemed to make you happy. But I didn't quiet grasp how so.
I dunno, looks like a tank to me ;)
The edges of the tree sharpen only because there is an object behind it which is of interest, the actual internal details of the tree diminish. With just the outline fo the tree, its easier for a neural network to recognize the tank as there are fewer details that belong to the 'background' classification.
Here are a few more images, I probably need to work on its ability to detect a tank in a sandstorm.
OK, fixed it by adding a setting to allow the amplification of low contrast images -
A fixation on tanks. I see...
My interest died here.
I was expecting a strong German/slavic/nordic accent. Couldn't detect bugger all. ;)