Well, I got the template matching going this evening. It works great. It is very forgiving for size and angle as can be seen by the images below. The thing I really need to do next is get a somewhat bigger sample taken of cars pulling up to a fixed location so I can better characterize the size of the plate in the image over a fairly large range of cars. The image of the pickup plates were taken further away than for the car. There was variability in both angle and size for all the plates, but the same template set was able to find them all very well. I have a few cleanup things I need to do before I move on which include some speedup work and some precsion improvements. I am searching in rotation right now, but not in scale. After I see the more controlled sample set, I will know whether I need to add scale. After that, I will move on to segmenting the individual characters.
Day: July 21, 2012
Day 335 of 1000
Today, I was able to build an “on-the-fly” template that handles ±5º of rotation. I just have the template search complete, too, but I do not think I will get it completed before the end of the evening. Here is a GIF animation of all the patterns for the template search described yesterday.