Robust Estimation of a Multi-camera System Motion Parameters Using Inter-camera Mounting Parameters

Accurate Indoor 3D models could be a valuable asset for many applications such as building reconstruction, rescue operations, and Building Information Management (BIM). Through image processing, 3D models can be generated from high quality images captured by off-the-shelf digital cameras. To acquire redundant data and produce real scale models, a multi-camera system can be used. This paper presents a methodology for robust estimation of motion parameters of a multi-camera system. First, the interior orientation parameters and mounting parameters among the cameras are estimated through a single-step procedure using images covering a calibration test field. Then, synchronized images are taken at a given time interval while the system is moving through building corridors. In the next step, features are extracted and matched between all possible intra-epoch and inter-epoch image pairs. False matches across Epipolar lines are filtered out in two steps: a) using the known mounting parameters for the intra-epoch matches and b) while estimating inter-epoch Relative Orientation (RO) parameters. In the proposed methodology, the system rotation (R) and translation vector (r) between successive epochs are estimated through rotation and translation compatibility constraints that involve the mounting parameters among the individual cameras on board the system. Knowing the system motion parameters (R and r), outliers along and across Epipolar lines can be detected. Preliminary results have shown that the system motion parameters are reliably estimated using the proposed methodology.