Body-Eye Calibration and Multiple-view Uncalibrated Reconstruction

Leo Reyes

This presentation has two parts. The first part of the presentation concerns the problem of autonomous robotic calibration (the computation of the relationships between the sensors and effectors of a robot). Specifically, the problem of eye-to-body calibration is discussed and compared to the classica Hand-Eye calibration. A new method proposed to compute this transformation is presented and some experimental results are shown.

The second part of this presentation concerns the problem of multiple view uncalibrated reconstruction using Bundle Adjustment (BA). Traditionally, BA has been used to make projective reconstructions of cameras and points. Currently, only few works are available to treat the problem of simultaneous reconstruction of various primitives using this approach. One of the contributions of this work is that we extended BA to reconstruct points, lines, quadrics, degenerate quadrics and plane conics. Also, another significant point that was discovered is that the classic cost function used to solve this problem was not good enough since the topology of the quadrics was sometimes lost. Therefore we had to modify the classic cost function in order to keep the topology of the quadrics. Finally another contribution of this work is the presentation of a novel algorithm for the correction of conic outlines in n-views.

Maintained by Philippos Mordohai