From: Chris Russ Subject: Re: Object/difference detection in static scene? Date: 12 Aug 1999 00:00:00 GMT Message-ID: <37B31C44.D3A64B09@aol.com> Content-Transfer-Encoding: 7bit References: X-Accept-Language: en X-Notice: should be reported to postmaster@ibm.net Content-Type: text/plain; charset=us-ascii; x-mac-type="54455854"; x-mac-creator="4D4F5353" X-Complaints-To: postmaster@ibm.net X-Trace: 12 Aug 1999 19:14:12 GMT, 32.100.213.48 Organization: Reindeer Games, Inc. MIME-Version: 1.0 Reply-To: jcr6@aol.com Newsgroups: sci.image.processing You know, you could catch the boats when they move --- if you can take the difference between frames a reasonable period of time apart (say 1 second?). The background will be completely removed and there will be a negative area where the boat was and a positive area where the boat has moved. By histogramming the difference image and looking at the % of the image above some cutoff, you can automatically detect motion... Motion that any boat would cause (or seagull or other biological). Not to mention violent weather. ;-) Glen Harris wrote: > I'm designing a system which proceses a series of images from a fixed > camera. The point is to detect any boats in an area of water. If the > system finds one or more, the image is flagged for manual analysis, > otherwise it is ignored. Processing time is not really a problem. > several images/second would be fine, although it could run overnight > and the flagged images be waiting next morning. > > Now, thinking about it, there are several simple ways of doing this, and > several problems. Firstly, there should be a known "empty" scene, or a > few of them, from which to calculate the difference. Since lighting > conditions change throughout the day/year, plus clouds, etc, the system > should probably greyscale the image, then do some average brightness > correction before analysis. > > The actual analysis, I thought, might start out by subtracting the > "empty" image from the unnkown image. This would give a pretty blask > image, with any objects as grey blobs, which can be detected. > > This is a very simplistic and probably inefficient system, and I'm > sure there's better ones out there, but it gives you an idea of what > I'm after. BTW, code preferred, psuedocode nice, detailed algorythm/ > explanation OK, pointer to Graphics volume 3 pp 345 is almost > useless. > > Suggestions, anyone? > > Thanks, glen.