Conjugate Gradient Schemes for Motion Estimation in Image Sequences,
Abstract
This report describes methods for the estimation of local motion in a sequence of images. The methods are based on a global minimization of a relevant error criterion using a conjugate gradient technique. The expansion of the displacement field is carried out with the use of basis functions according to a Fourier transform or a cosine sine transform in space and time. These representations automatically result in a continuous displacement field where subpixel displacements can be detected. The error criterion can be based on a displacement vector for each pixel, but also on linearized versions in vectorial and scalar form, depending on the desired trade off between computation time and accuracy.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 01, 1997
- Accession Number
- ADA326106
Entities
People
- A. P. Berkhoff
Organizations
- Netherlands Organisation for Applied Scientific Research