I have implemented the condensation algorithm in order to track a moving object in image streams. If the moving object is simple, the object's state can be represented only by the coordinates of the center of its bounding box.
However, if the object is curved, it may be subjected to transformations that change its shape during movement. In this case, the object's state cannot be described by a single point (the center of its bounding box), but it would make more sense to use different points on the curve. How to choose the points which represent the state of the object?