An efficient shortest triangle paths algorithm applied to multi-camera self-calibration

摘要

We propose a novel minimum uncertainty approach to relative pose selection for multi-camera self-calibration. We show how this discrete global optimization problem can be expressed as a shortest triangle paths problem. For the latter, we present an efficient algorithm and prove its correctness. It has several advantages compared to a similar approach of Vergés-Llahí, Moldovan and Wada. In quantitative experiments on publically available data, we show that our relative pose selection method provides improvements compared to naive, random and greedy selection, without and with subsequent bundle adjustment.