Automatic Calibration of Multiple Fisheye Cameras: Trajectory based Extrinsic Calibration of a Multicamera Rig

Abstract

This thesis presents a new technique for automatic extrinsic calibration of fisheye cameras on boats. In contrast to most of the existing methods that require overlapping field of view, our technique does not, which opens up for completely free placement of cameras. The method selected for this work was trajectory based calibration, meaning that the fixed relative pose between the cameras can be determined from the trajectories of the cameras. For development, Gazebo VRX simulator was used to generate the data for calibration, and KITTI odometry to evaluate algorithms. The method for generating the trajectories was visual inertial odometry and ground truth trajectories were extracted from the simulator. The calibration method was selected based on the availability of sensors onboard boats, for the objective to generate accurate trajectories. It was found that the developed trajectory based calibration of relative pose between cameras is working, given an accurate trajectory. The method was independent of which algorithm is used to obtain the trajectory. However, there were difficulties getting visual inertial odometry alone to generate trajectories accurate enough for the simulated data sets.