I have a question regarding Template matching for object recognition in optical remote sensing data. My goal is to identify lamp poles from optical data. It is a little tricky to do so because the lamp poles are only recognizable by their shadows and not the object itself. Also there are problems with the translation and the orientation of the objects in the image. I would like to ask what are the recommended pre-processing steps such as, smoothing, sharpening, contrast stretching and so on, in order to enhance the performance of the template matching? and also what is the most suitable similarity measure to use? Currently I am considering NCC calculated in the Fourier domain. Are there more robust estimates?
Thanks a lot in advance for your response, Sina
Have you had any luck with this?
I am also working on a similar application and I was wondering if you got any help here.