Detecting Light Pole

edit

What you could do is basically two approaches

1) Creating a model of a general light pole structure, then perform a detection using that model. Techniques that are usefull are SVM, random forest, cascade and naive bayes classifiers. You can find more information on them in the machine learning package of openCV.

2) If you know the exact appearance, you could use a technique called feature point matching. You first search for interesting points in the general model image, then try to find those interesting points again in the scene. Finishing the detection by using a feature point matcher, that tries to find identical feature points. You should take a look at the features2D package of openCV for that.

Always take into account that object detection is not a straightforward problem. There is no general solution, only good working solutions for specific situations. This makes that you need to try out different approaches in order to find a good match for your problem.

If your pole stands out with respect to background, you may be able to perform simple segmentation, identify vertical objects (height >> width, and angle at 90 degrees with respect to ground) and then, perform an erosion on that object to get a line.