Abstract
Recently, the problem of estimating motion fields from image sequences is becoming important for root vision and so on. We already proposed a method for estimating an entire motion-vector field from a given set of image-sequence data. In the method two different methods, the matching method and the gradient method are realized and fused by the use of learning and function approximation abilities of neural networks. The purpose of this study is to investigate the performance of the proposed method through experiments in several situations. It is shown that the proposed fusion is effective and makes it possible to estimate motion fields more accurately.
