Recognition of magazine pages
For marketing research purposes it is interesting to know how long each page of a magazine is read. Therefore we developed a prototype to recognize magazine pages for a customer. Scanned or digital pages were processed and a model was created to recognize magazine pages from mobile phone or USB Webcam video recordings. By using a portable camera with wide angle lens which can be put into a breast pocket, the prototype is able to "read" the pages without changing reading behaviour of the test subject. A temporal resolution of 500 ms (= two page recognitions per second) with an accuracy of more than 99.5% was achieved in tests.
The video shows a demonstration on eMedia No. 12 from 2010/06/18. At the bottom right the page is shown. The page is always correct for this video. If recognition is not possible for a frame, no page is shown. We have also ported this demo to an Android Tablet in 2012 where it runs at around five frames per second using only local computational resources.