People counting by learning their appearance in a multi-view camera environment