I work for a small startup downtown. We don’t have any designated parking, just street parking with meters. This can be quite an annoying situation and my coworkers and I end up getting a lot of tickets. So I want to build a meter maid detector.
I have a raspberry pi with a camera I’m using for the project. I’m trying to use openCV to recognize the meter maid when he comes by and send an alert. The maid meter, 99% of the time, wears a florescent yellow-green vest with reflective strips. I have seen them a few times wearing just a blue shirt. So my plan is to recognize that vest and maybe the blue shirt version as well.
My early attempts have mainly involved converting the image to HSV and trying to see how many pixels are in the range similar to the vest. I was however getting a lot of false positives especially near dawn and dusk. I figure I should probably add in background subtraction so I don’t have to factor in the non-moving part of the scene. I’m thinking there is probably a better solution than trying to compare pixels in a range. I’ve ran into mean shift tracking and histogram matching, that sound promising, but I’m sure there’s many other methods. I want to be able to detect him at different ranges and in different lighting conditions. Any ideas what would work best for this type of object recognition?
