Local iterative DLT for interval-valued stereo calibration and triangulation uncertainty bounding in 3D biological form reconstruction

Abstract

The use of stereo vision for 3D biological data gathering in the field is affected by constrains in the position of the cameras, the quality of the optical elements and the numerical algorithms for calibration and matching. A procedure for bounding the 3D errors within an uncertainty volume is also lacking. In this work, this is solved by implementing the whole set of computations, including calibration and triangulation, with interval data. This is in contrast with previous works that rely on Direct Linear Transform (DLT) as a camera model. To keep better with real lens aberrations, a local iterative modification is proposed that provides an on-demand set of calibration parameters for each 3D point, comprising the nearest ones in 3D space. In this way, the estimated camera parameters are closely related with camera aberrations at the lens area through which that 3D point is imaged. We use real data from previous works in related research areas to judge whether our approach improves the accuracy of other crisp and intervalvalued estimations without degrading the precision, and conclude that the new technique is able to improve the uncertainty volumes in a wide range of practical cases.