Path: utzoo!utgpu!news-server.csri.toronto.edu!cs.utexas.edu!usc!zaphod.mps.ohio-state.edu!uakari.primate.wisc.edu!umriscc!maverick.ksu.ksu.edu!hoss!typhoon!willdye From: willdye@typhoon.unl.edu Newsgroups: comp.ai Subject: Re: Shooting pigeons Message-ID: Date: 22 Mar 91 17:10:16 GMT References: <4754@syma.sussex.ac.uk> Sender: news@hoss.unl.edu (Network News Administer) Organization: University of Nebraska - Lincoln Lines: 46 I mailed the following letter to him, but then I thought the rest of you might want to see what I wrote, so here's a copy. ----------------------------------------------------------------------------- To: peterhi@syma.sussex.ac.uk Subject: Re: Shooting pigeons Newsgroups: comp.ai References: <4754@syma.sussex.ac.uk> In comp.ai you write: >In a >moment of pure insanity I submitted my project as to "Design and build a >system that that shoot clay pigeon" the basis of the design is that everything >can be done with a single camera and no more information than a shooter would >have. Start grovelling at the professor (or whatever they call 'em in England) to accept 'signifigant effort' instead of a working system. For YEARS now the military has been developing robotic systems that don't have to do much more than what you're suggesting. It's been said that if anyone developed a system capable of playing ping-pong (table tennis), it would be locked up and classified. You have two major problems. One is a computer vision problem of identifying and tracking the target IN REAL TIME. This is non-trivial, even when you have artificial constraints put in to make sure the signal-to-noise ratio is high. The second problem is building the mechanics necessary to move the shotgun ACCURATELY and QUICKLY to the desired position. This is also non-trivial, and also quite a different problem from the software-intensive vision problem. I'm not saying don't do it. In fact, I'd LOVE to work on something like this. I'm just saying start small and don't have illusions about the task before you. I suggest starting with a simplified software simulation. Have one computer (or a camera) send a video picture of a black ball moving in a straight line against a white background. See if you can keep crosshairs on the ball. Then use a ball with a non-straight flight path, and try to LEAD the target in real time. It may help to look up some basic references to radar tracking algorithms, stuff like aquisition gates (large area around the target), tracking gate (small area around target), and turn gates (in-between sized gate). Don't try a fancy system with a high signal-to-noise ratio until you've got the simple system working. The S/N ratio problem can never truly be 'solved', we just get better and better at it. Good luck. Please tell me how it worked out. William L. Dye willdye@typhoon.unl.edu