Merging multiple stereo surface maps through camera self-calibration

A method for estimating surfaces from a sequence of stereo images is described. It is assumed that the three-dimensional scene is stationary, and that the cameras can be aimed at new scene locations for image acquisition. For some applications, these individual stereo maps need to be combined into a central, composite representation. The author describes a method for incrementally merging such maps by allowing small perturbations in calibrated imaging parameters to achieve an optimum fit among the estimated surfaces. This is equivalent to integrating the processes of surface estimation and adaptive self-calibration. The method is tested with a machine-vision system which automatically aims and focuses two cameras at scene targets, acquires images, and constructs a composite surface estimate