Sunday, January 24, 2010

Comparative data on machine vision performance

When we’re “selling” machine vision to management one of the arguments is the superior accuracy and repeatability as compared to human inspection. Unfortunately though, data on this is hard to come by. I suppose that most end-users are reluctant to give away much information that could come back to haunt them, or is it that the systems aren’t as good as we’d like to believe?

Whatever the reason, Andy Wilson is helping to redress this data deficiency. Writing in the December 2009 issue of Vision Systems Design, (“
Vision system speeds egg crack detection,”) he gives a fascinating description of a vision system under development at the US Agricultural Research Service. The real nugget of gold comes in the very last paragraph where Andy provides data – yes, real numbers! – comparing the performance of the vision system to that of human inspectors.

I’m not going to steal Andy’s thunder by reproducing the numbers here – you’ll have to read his article for that – but what I will say is that it rather undermines that oft-repeated assertion that human inspection is only 80% effective.

One last thought: is egg inspection yet another example of agri-vision? (See my previous post if you don’t know what I’m on about.)

No comments: