Robust position measurement in visual eigenspace

A survey and analysis of visual measurement of camera and object position in visual sub-space (eigenspace) is provided leading to several improvements to existing methods, as well as new approaches. Specifically, novel techniques were developed to allow robust measurements in the presence of occlusions and other dynamic scene changes which is known to be a significant challenge in pose measurement methods for important applications such as visual servoing, autonomous robotics in manufacturing and tele-robotics including aerospace, medical operations and others.; Local image information is shown to retain positional information in unoccluded regions that can then be used to determine position in the presence of significant dynamic occlusion. Local information is also shown to be more prone to ambiguity errors due to a lack of salient positional information. A subsectioning and recombination strategy is developed that features the advantages of local eigenspace in-dependence for robustness to occlusion while maintaining the inherent resistance to ambiguity available from global eigenspace analysis. This is achieved by computing modified global projections, while excluding information from occluded sections. A new method for occlusion detection using an eigenspace reconstruction error measure is also developed and evaluated.; A wide variety of experimental measurements are provided to demonstrate the performance of the new methods using an accurate XYZ platform and CCD cameras with metallic, machined parts. Experimental measurements are also performed to demonstrate improvements for eigenspace position accuracy through the use of multiple cameras. Several techniques are employed to combine and fuse multiple images from decoupled cameras whereby cameras are used for determining position in different directions to improve accuracy. Subsequently, multiple cameras are applied to achieving three dimensional translational position measurements in visual subspace.