Machine vision in surveillance

In the main I restrict this blog to industrial applications of machine vision, but on several occasions I’ve mentioned that security is a “target-rich environment” for developers of vision technology, so let’s take a short detour in that direction.

Surveillance relies to a great extent on the use of “pan-tilt-zoom” (PTZ) cameras. This results in lots of overlapping images with different scale factors, which are difficult to stitch together into a single scene. Researchers Marcos Nieto and Luis Salgado have been working on this problem and in “Automatic video mosaicing for surveillance using vanishing points” (SPIE web site, March 17th, 2010,) they set out a rather interesting approach.

As the title of paper suggests, their trick is to use vanishing points – those lines that disappear into the distance, meeting at some far point – as a way of registering disparate images. And guess what – it seems to work! (Check the paper to see the images.)

As a huge fan of the cult 1971 movie, “Vanishing Point,: I can’t help but be attracted to this method, even though Nieto and Salgado seem not to using a Dodge Challenger in their work. Check out this imdb link if you don’t know what I’m rambling about.