Abstract: The position of a next object is estimated efficiently according to the observed value of physical quantity related to the object in an actual environment. An image processing apparatus extracts a first object image of state prediction from a first image and identifies the type of the first object. A model appropriate to the type of the object of state prediction is merged with the first object image and a feature point of the first object image is associated with the model. A plurality of motion sequences are selected from a finite state sequence corresponding to the type of the first object according to the position of the first object and the initial observed state.