next up previous index
Next: Variational Approach Up: Low Level MRF Models Previous: Texture Segmentation

2.5

Optical Flow

 

Optical flow is the velocity field in the image plane caused by the motion of the observer and objects in the scene. It contains important information about cues for region and boundary segmentation, shape recovery, and so on. This section describes methods for computing optical flow and motion from a sequence of time-varying images.

Two major paradigms exist for determining visual motion: feature-based ( e.g. edge-based) [Nagel 1983 ; Paquin and Dubios 1983 ; Bouthemy 1989] and gradient-based [Horn and Schunck 1981]. In the former paradigm, features are extracted from the sequence of images, matched between two neighboring frames and tracked over time. This gives a sparse flow of which the information is available at the sparse set of the extracted image features. In the gradient-based paradigm, the flow is recovered based on local spatial-temporal changes in image intensity. This gives a dense flow field whose values are available throughout the image plane. The focus of this section will be on finding flow field using gradient-based methods.