Extrinsic calibration of a multi-Kinect camera scanning passage for measuring functional traits in dairy cows

Camera based systems in dairy cattle have been intensively studied over the last years. Past studies have concentrated on single camera systems with a limited range of applications. Here the development of a camera system comprising multiple 3D cameras (six Microsoft Kinect cameras) for monitoring purposes in dairy cows is presented. A recording unit was constructed, and software for recording, synchronising, sorting and segmenting images, and transforming the 3D data in a joint coordinate system was implemented. The latter is called extrinsic calibration and was dealt with in this study using a 3D calibration object. The internal estimations of acceleration data and coefficients for the parametrised plane describing the floor of the scenery obtained by the Kinect camera were tested and discussed with regard to their usage in the calibration process. The presented approach accurately determined whether a Kinect device was a side view camera or one of two types of top view cameras (accuracy 92.9%) and specified its mounting position in the system (accuracy 95.8%). Furthermore, rotations and translations between the cameras and a reference coordinate system were estimated from 3D point clouds of the calibration object. For cameras in upright position (side view cameras) the rotational accuracies were satisfactory, showing 4.4° (±2.9°) deviation from the reference. For the cameras in top view position larger errors (25.5° ± 6.4°) were measured due to them being mounted in an inclined way. Nevertheless, a basis for registration methods was achieved.