This paper presents details of techniques for projecting image data from microscope views to perspective views or model spaces. These software tools are used to assist human operators with visualisation and navigation in semi-autonomous tasks performed at the micro-scale. The problems of applying computer vision techniques at this scale are highlighted. The solution ensures minimal reliance on prior calibration, and overcomes the problems of finding correspondences over scale differences by accumulating data in image mosaics.

A technique suitable for tracking rigid motion in microscope image sequences is presented. Microscope images have particular problems: notably sparsity of features and ambiguity of motion. The method searches for point correspondences but concentrates on random transitory features rather than regular geometry. The paper highlights the problems that must be faced and details the work that has been implemented to combine robustness with efficiency. The output of the search algorithm are sets of correspondences and the inter-image transformations.