Automated Surface Extraction in Real Time Photogrammetry

Considerable research attention has been given to close-range videometric systems for vision-based, three-dimensional object measurement over recent years. These systems have to demonstrate high speed, high accuracy capability and high flexibility to compete as a three-dimensional coordinate measurement tool. Especially in industries where instant feedback is required, such as production line or tracking applications which demand a process control response, realtime or near-real-time photogrammetric systems are needed. Automatic feature correspondence determination is the central problem in the development of such systems. This paper reports on the development of a multi-sensor digital close-range photogrammetric (videometric) system for the determination of surface contours of featureless and targetless objects. The use of structured light projection in such systems is a common approach for providing necessary surface texture to support image matching. A new strategy which involves the hierarchical use of projected non-repeating patterns to enable fast, unambiguous image correspondences to be established in an initial coarse surface extraction has been employed in this system. The final, refined surface contours are then recovered via feature-based matching to a projected ‘random’ pattern. The paper first describes the non-repeating pattern strategy and the automated image matching approach for surface reconstruction. This is followed by a description of the multi-sensor vision metrology system built to facilitate the experimental work of this research. The integration of the approaches into the system is discussed and experimental measurements conducted are summarised.