Faculty of Physics and Astronomy University of Heidelberg

This thesis presents a successful novel approach to the challenging problem of model selection in motion estimation from sequences of images. The concept is based on the brightness change constraint which is discussed formally and concisely such that new light is cast on the properties of local parametric optical flow models. These models give rise to parameter estimation problems with highly correlated errors in variables (EIV). Regression is hence performed in a statistically suitable way by means of equilibrated total least squares (ETLS). The author relates the probability distribution function of regression residuals to the EIV model. Under the assumption that parameter estimates are deterministic, these residuals can be decorrelated. The author suggests in the assessment of motion models to test the decorrelated residuals to stem from the distribution which is theoretically imposed by the EIV model. It is demonstrated in simulation that this concept shows good performance in the exclusion of inappropriate models and yields promising results in model complexity control and in the identification of motion discontinuities. Different statistical distribution tests are involved and compared in the simulation. This thesis is the first application of statistical residual analysis in motion estimation.