Calibration of a multi-camera system

Mehrkamerasystem i3mainz, CC BY SA 4.0

A 360° multi-camera system was developed for measuring shafts in mining, which was calibrated on the i3mainz. With the help of a simulated calibration, a suitable calibration field was set up and a suitable process was developed with which the internal and relative orientation of the cameras could be determined.


A 360° camera was developed by DMT GmbH and DLR for measuring shafts in mining. This camera is mounted on the shaft elevator and is intended to record an overlapping image group of the shaft during a journey. A digital reconstruction of the shaft can then be created using these images and the 3D photogrammetry. In order for the system to deliver the most accurate results possible, it is necessary to calibrate the individual cameras. This calibration was carried out in the i3mainz laboratories. The required accuracy of the calibration is 1 cm for object points with a distance of approx. 4 meters.


The calibration of the multi-camera system consists of determining the internal orientation of the individual cameras and determining the relative orientation of the cameras to one another. Due to the relatively heavy system (approx. 40 kg) and the resulting limited mobility, a combined test field calibration procedure was used, in which the parameters of the inner and outer orientation are determined by an adjustment calculation.

When setting up the test field, several factors had to be taken into account. On the one hand, the object distances (distance from the camera to the shaft wall) during calibration should correspond to the distance during the recording. On the other hand, the cameras and lenses used impose certain requirements on the size and distribution of the targets. These factors are first examined in a simulated calibration in order to then create a real test field.

Due to the distribution of the 8 cameras over the full horizontal picture angle of 360°, the test field was also set up spatially around different camera positions. The size of the calibration field is approx. 8 m x 6 m x 4 m (length, width, height) and consists of coded and uncoded photogrammetric targets. The distance between the target marks is about 40-80 cm. The position of the target marks was determined in a network measurement with a precision tachymeter, in which care was taken to ensure that favorable cutting conditions were present at the target marks.

To calibrate the camera, shots had to be taken in different positions. In total there are 18 positions on two levels. At each of the 18 positions, the camera was rotated once around its own vertical axis in 12 steps. The evaluation of the 216 recorded images per camera takes place in a photogrammetric bundle block adjustment, in which the position and orientation of the image as well as the inner orientation of each camera is calculated.


When determining the target position, an accuracy of 1.7 mm (RMS of the confidence ellipsoids) was achieved, which is sufficient to meet the required accuracy. The parameters of the inner orientation were determined directly in the course of the bundle block adjustment. The relative orientation is given in relation to camera 1 and was determined by averaging the external orientation of each camera. The calibration was checked by independent control recordings. The final accuracy was 7.6 mm (RMS of the 3D residual gaps) and thus meets the required accuracy. In test measurements, an increased calculation speed and improved accuracy of the camera system could be determined.