Expand this Topic clickable element to expand a topic
Skip to content
Optica Publishing Group

Computing motion using neural networks

Open Access Open Access

Abstract

Computing the motion or flow field is an illposed problem [Poggio et al. (1985)]. It can be regularized by the use of the smoothness assumption [Horn and Schunck (1981)]. The motion field should be the smoothest compatible with the data. However, these schemes perform badly in the presence of several objects moving in different directions. This major difficulty can be overcome by the introduction of binary line processes [similar to Geman and Geman (1984)] to model discontinuities in the flow field. If the local gradient between neighboring values of the flow field becomes too large, the appropriate line process is turned on, segmenting the velocity field at that location.

© 1987 Optical Society of America

PDF Article
More Like This
Efficiency of computation in neural networks

Yaser S. Abu-Mostafa
WK3 OSA Annual Meeting (FIO) 1985

Synapses that compute motion

Christof Koch
MT2 OSA Annual Meeting (FIO) 1987

Extracting 3-D curvatures from images of surfaces using a neural network

Sidney R. Lehky and Terrence J. Sejnowski
MH3 OSA Annual Meeting (FIO) 1987

Select as filters


Select Topics Cancel
© Copyright 2024 | Optica Publishing Group. All rights reserved, including rights for text and data mining and training of artificial technologies or similar technologies.