Tracking block of text on a page

asked 2017-02-24 12:49:21 -0600

tahme gravatar image

Hi, I'm having a hard time getting good results from tracking just a word or short phrase on a page of words. My basic workflow is to do OCR on the text and get boundbox of the section I want and then init a tracker in that area. As the page moves I update the tracker, but it quite easily skips around to another body of text. I'm wonder if there's any suggested algorithms or workflows for this kind of task (I'm currently getting best results from MIL). Trying to avoid doing constant detection because then the real-time speed is lost.
Any tips, much appreciated - thanks!

edit retag flag offensive close merge delete