| The goal of automatic surface reconstruction is to measure and interpret the object space surface with little or no human intervention. Current 'stereo matching' techniques first refine the stereo images in order to remove geometric and radiometric distortions between the images and provide distinct representations. Conjugate locations (parallax) are determined by similarity measures between these refined stereo images. However, complex geometric and radiometric properties and lack of information known prior to implementation compel these techniques to heavily constrain the space of possible solutions. This results in incorrect or over-generalized measurements of the physical world. A multi-scale, object space concept and subsequent strategy is proposed that more completely removes geometric distortions, decreases the need for unrealistic information prior to reconstruction, and integrates stereo matching and other surface reconstruction tasks in a common, goal-oriented environment: the object space. This strategy is implemented and compared with a traditional stereo matching strategy. The results show that even with rudimentary implementation techniques the proposed strategy increases the measurement robustness and interpretation richness of complex object space surfaces. |
