CitationDownload as .RIS
Emerald Group Publishing Limited
Copyright © 1998, MCB UP Limited
3D - just a matter of perspective
3D just a matter of perspective
This issue's theme is 3D space sensing and we cover various techniques that may be used for mapping the 3D profile of a surface.
Everyone knows that if you want to judge distance, to catch a ball for example, then you need two eyes and stereo vision so that your brain can work out the complex trigonometry and image cross-correlation needed to forma 3D map of your surroundings and the incoming missile.
Fortunately, nobody told this to my one-eyed friend at school who headed-up the cricket team. You can try it yourself; sure two eyes are better than one but you can still obtain a pretty good sense of spatial awareness with your monocular vision. This works because we can tell from a priori knowledge and general image understanding abilities that some objects are in front of others. Relative movement of the object or our head, and the fact that the ball is getting bigger, all enable us to reach out and capture it.
These sort of tricks are very easy for people and other life forms to perform, however, computers find this kind of thing very difficult indeed. I would not therefore suggest that we should try to mimic the human approach with our computers, however, I would suggest that we should challenge our conceptions about what is or is not possible. If my friend could become captain of cricket perhaps our existing vision systems can attain 3D sensing abilities.
Today's vision systems are pretty good at finding edges, which they can do to sub-pixel precision, and computers are also good at measuring time. So if an edge moves past us we can measure its speed and if we view it from somewhere closer than infinity the edges closest to us will appear to move the fastest. This means that the relative speed of various image components can be matched more or less directly into distance away from thecamera.
This approach may be old hat and if someone is working on it I would be pleased to hear from them; however, the point of this illustration is that if we alter our perspective from one of trying to do things with computers that are hard, such as image cross-correlation, to instead doing operations that computers are good at, such as detecting edges and measuring time, then we may find that we already have an easy solution to a difficult problem.
The next generation of advanced automation and robotics is going to need much more in the way of smart sensing abilities and a part of that need will be for useful 3D sensors. However, 3D sensors are still something of a speciality item with a few notable exceptions. For example, is it possible to buy a camera without auto-focus? Perhaps soon we will all have scanners front and rear to help us park our cars? The opportunities are great but we may need to change our perspective on the problem.