Vector field approximation by model inclusive learning of neural networks
ICANN'07 Proceedings of the 17th international conference on Artificial neural networks
Hi-index | 0.00 |
The problem of estimating motion fields from image se- quences is essential for robot vision and so on. This paper discusses a method for estimating an entire continuous motion-vector field from a given set of image-sequence data. One promising method to realize accurate and efficient estimations is to fuse different estimation methods. We propose a neural network-based method to estimate motion-vector fields. The proposed method fuses two conventional methods, the correlation method and the differential method by model inclusive learning, which enables approximation results to possess inherent property of vector fields. It is shown through experiments that the proposed method makes it possible to estimate motion fields more accurately.