best approach to detect different objects real-time

edit

There isn't a single technique out there in OpenCV that will give you a multiple class detector. Your best approach is to build two detectors, for each seperate class and combine them together in a search for candidates in order to detect both classes.

If the object remains the same, using techniques like SIFT and SURF to define feature points and match them is a possibility.

If however intra-class variation exists, you could try to build a model for each class, using the cascade classifier approach.

Give us more detail about your exact needs.