Backgroung Oriented Schlieren (BOS) is a Schlieren-like method that can be used as a 3D quantitative method for measuring density in 3D flows. This can be done providing that there are enough points of view, at least 10, and that an accurate calibration of the multi-camera bench is available. We present here a study of multi-camera calibration for 3DBOS. We use a simple printed flat plate with a dotted regular pattern as calibration body (CB). The main issues are the workload, because there are several hundred of images, and the blurring effect. The blur affects the images of the CB when it is positioned in the common part of the fields of views of the cameras. It can not be avoided because of illumination limitation. We present practical solutions and results for both issues. A GPU implementation allows to complete the calibration of a system with 12 cameras in less than 30mn. We also propose an uncertainty analysis using the jackknife method. It is shown that the uncertainty of the estimation of the horizontal field of view (HFoV) is 0.04°. Moreover, HFoV is nearly identical for all cameras and quasi equal to the theoretical value.