The majority of visual attention models is based on the concept of saliency map, a two-dimensional map that encodes the saliency of objects in the surrounding world. We attempt a short review of current implementations and present our first thoughts on extending the classical saliency-based model by using motion and prior knowledge.