Suggestion about method for recognition based on machine learning?

I am going to develop a traffic sign recognition based on machine learning (SVM). I just have to work at the recognition part (NOT in segmentation and detection), So I will get the classification (shape information from the previous work)

What I want to ask:

1. What are the best Features that I can use for recognize the traffic sign (despite shape and colours)? Because I will use PCA for features reduction, so I think I need more that 2 features (not only shape and colours)?
2. I heard that HOG Features can handle the rotations of the sign, do I still have to rotate my sign while I am using HOG features??
3. Do u have another suggestion what method I have to consider about Traffic Sign Recognition Part (For Feature extractor, SVM, tracking the object, rotation handle, occlusion of the object)??
4. Do u have suggestion about web, book, tutorial, library function in MATLAB that I can refer to study all about this?

Thankss