An experimental investigation of estimation approaches for optical flow fields

Various approaches have been suggested to solve the correspondence problem for image sequences. This chapter discusses two basically different approaches for the estimation of optical flow vector fields in order to clarify their advantages and disadvantages as well as the relation between them. The first approach is based on the extraction and interframe match of features. Several features are investigated using real-world images. The performance of the approach involving features is evaluated. In distinction to feature-based approaches, one may directly evaluate the spatio-temporal variation of the picture function which results in a differential or gradient-based approach. This gradient-based approach has been supplemented by a smoothness requirements on the optical flow field, in which the estimation of the optical flow field formulated as an optimization problem resulting in a system of partial differential equations for the optical flow field. Subsequent investigations show that s ignificant contributions for the determination of the optical flow field could be expected among others from gray value structures which correspond to the image locations evaluated in the feature-based approach. Results from both approaches obtained using the same input data are compared.