Topological Auto-Calibration of Central Imaging Sensors

摘要: This paper departs from traditional calibration in the sense that pixels forming camera have a completely unknown topology. Previous works shown statistical properties of natural scenes, and uniform motion both translation rotation, allow determining topology central [10,6]. Here we show there is quasi-linear relationship between time-correlation angular inter-pixel distance, considering small angles simple scenario encompassing one bright light on dark background. The reconstruction algorithm therefore based correlating time series (pixel streams) acquired by moving camera. Correlations are converted to distances using fixed linear transformation. Distances finally embedded plane manifold learning methodology, namely Isomap. Experiments real datasets been conducted theoretical derivations accurate for considered scenario.