Kinect sensors network calibration in controlled environment based on semantic information

Calibrating camera networks is of importance in applications where object of interest must be analyzed either from multiple views or as a single 3D entity. The selected cali-bration procedure mostly depends on the required accuracy and usually involves a dedicated calibration target named calibration marker. This is a pattern printed on 2D sheet or can be a 3D object with dedicated shape and known dimensions. Using such markers however, might not be convenient in practice. Such situations arise when the calibration procedure requires special expertise or time consuming frequent re-calibration. In such cases the dedicated calibration marker can be substituted with objects which are part of the surrounding environment. Usually such objects contain some semantic meaning, e.g. walls, tables etc. In this paper, we propose an approach for calibrating Kinect sensors network using objects with semantic meaning, part of a controlled environment. Our goal is to avoid the necessity for dedicated markers, which complicates the calibration procedure and to deliver simple tools to the end user. The experimental re-sults, show that the proposed approach for semantic calibration delivers high accuracy results, similar to the results of state of the art algorithms for Kinect sensors network calibration.