Interactive semantic camera coverage determination using 3D floorplans

Emerging applications like smart environments and observation systems can benefit significantly by making use of semantics provided by floorplan of the environment. Cameras in the system can be calibrated with respect to the floorplan which allows their coverage to be automatically labeled with semantically meaningful regions like walls, floors and doors. Calibration with respect to floorplan requires establishing correspondences between 3D features (corners, edges) and their 2D images. In the absence of reliable methods to establish these correspondences automatically, a human is often involved in the loop. This task of specifying correspondences manually is both tedious and prone to specifying redundant correspondences. This work attempts to make the task easier and intuitive for the user by identifying which 3D features are best to specify correspondence for. We propose a new measure to rank 3D features according to the amount of information they can provide for calibration, and develop a smart algorithm to determine the importance of 3D features based on this measure. This leads to an intuitive interface to guide the user in correspondence specification task. Results on simulated and real data look promising and closely follow the ideal behavior.