Robust Optic Flow Computation with Support Vector Regression

Differential methods for optic flow estimation suffer from some well know theoretical and practical limitations such as the "aperture problem", sensitivity to noise, intensity discontinuity, etc. This paper presents a new locally robust method to solve the optic flow constraint (OFC). Here, the OFC is formulated as a robust linear regression problem resolved by support vector machines. Outliers are automatically identified as support vectors and are removed with a gradually decreased insensitive e-margin. The performance of our approach is studied and compared with other recent methods.