In-car machine vision

Smart cars – not the tiny box on wheels but vehicles that can “think” for themselves, are coming, but technical details about the systems involved are often hard to unearth. Lifting the cover on that just a fraction is “Subaru Gives EyeSight to the Distracted,” published on the Wards Auto website, September 14^th, 2012.

This discusses a two camera solution implemented by Subaru. There’s a photo here and a moderately impressive video here, but what intrigued me were some of the technical details.

For instance, we are told that it, “functions by identifying vertical surfaces and contrast”. Edge detection in a horizontal direction only?

We are also told, “The cameras scan the road ahead every 0.10 seconds, and the distance to the lead vehicle is calculated based on the lateral distance between the two images.” That seems a rather low frame rate, and one wonders if distance calculation could be improved by mounting the cameras in the “A” pillars rather than behind the mirror?

And last, “The system needs about 2.5 seconds to react when an object comes into range.” Two and half seconds? That’s an eternity in the machine vision world.

Of course, auto makers have many constraints to deal with – price and interior packaging being two of the biggest – and since testing takes so much time they are not likely to be at the cutting edge of camera technology. But with all that said, this article gives what I found to be an interesting glimpse of where one machine vision “niche” (if the automotive sector is a niche,) is headed.