In this paper, a new framework for semi-automatic feature extraction is developed and applied to highway extraction and vehicle detection from multiple-frame aerial photographs. The basis of the new framework is a geometric deformable model. This model refers to the minimization of an objective function that connects the optimization problem with the propagation of regular curves. The implementation of curve propagation is based on level set theory. Utilizing implicit representation of two-dimensional curve, level set implementation is capable of dealing with topological changes naturally, and the output is independent of the position of the initial curves. In the previous study of the geometric active model, only boundary information is incorporated into the curve propagation process. Leakage is often inevitable wherever weak edge information exists. In this research, region-based information is added into the geometric active contour model and behaves as a constraint. Thereby, the new proposed method has the ability to integrate boundary and region-based information during the curve propagation and successfully solves the leakage problem. Several practical issues such as seed point selection and propagation are also discussed during the application of highway boundary extraction using this method from aerial photographs covering large areas.
展开▼
