Multi-primitive hierarchical stereo matching for 3-D measurement of road structures

A multi-primitive hierarchical (MPI) system for extracting 3-D information about the environment from stereo photographs taken by a mobile vehicle is presented. Low-level (edgels) and high-level (areas) features have been used to perform a fast and accurate analysis of the scene. The overall system integrates two commercial CCD cameras and a standard personal computer into an architecture that has been installed inside a mobile vehicle. The system is controlled through a single computer program called Prometheus. Camera calibration, image matching and 3D reconstruction are performed, as well as `ad hoc´ feature extraction tasks, peculiar to the specific problem