Telecentric imaging revisited

Never forget that in machine vision you’re not interested in artistic, pretty images: what you want are images that you (or your computer,) can analyze. This means that perspective, far from being a useful way of portraying depth in a 2D image, is a royal pain-in-the-ass.

Imagine a camera mounted on the front of a locomotive. The image from the camera will be of two rails that appear to be closer together the further away from the camera they are. Of course, you and I know the rails are parallel, but your poor PC doesn’t. So how do you measure the rail width in your image?

You can’t. Unless you use a telecentric lens. To quote Nicholas James of Edmund Optics, writing in Vision & Sensors, (Sensor Standards Require Superior Optics, April 29, 2009,) “In general terms, a telecentric lens is one that eliminates object-space parallax, or perspective,”.

If you want to get a better understanding of what a telecentric lens can do for you, follow the link and read the paper. Read it carefully because there’s a lot of detailed technical information, including a brief discussion of MTF and the diffraction limit. (Nicholas; is that what they call “burying the lead”?)