Adjustable linear models for optic flow based obstacle avoidance

An original framework to recover the first-order spatial description of the optic flow is proposed. The approach is based on recursive filtering, and uses a set of linear models that dynamically adjust their properties on the basis of context information. These models are inspired by the experimental evidence about motion analysis in biological systems. By checking the presence of these models in the optic flow through a multiple model Kalman Filter, it is possible to compute the coefficients of the affine description and to use this information for estimating the motion of the observer as well as the three-dimensional orientation of the surfaces in some points of interest in the scene. In order to systematically validate the approach, a set of benchmarking sequences is used, and, finally, the proposed algorithm is successfully applied in real-world automotive situations.