Abstract:
Projector-camera systems are gaining popularity in wide variety of applications like mobile
phones, interactive presentations, flight simulators, games and augmented reality applications.
Whenever projector is installed in a new setup, geometric calibration is performed to visually
correct the projected contents. Geometric calibration techniques are classified as active online,
active offline, passive online and passive offline. For applications where projection or display
surface is dynamic or deformable, online geometric calibration is required to adapt the surface
changes, and correct the projected imagery geometrically. This is because offline techniques
cannot tackle on the fly changes of projector position or display surface. Active online
techniques use imperceptible patterns along with high speed capturing cameras and Digital Light
Processing projectors. These devices are expensive and imperceptible patterns reduce dynamic
interval of projected contents. Passive online techniques require prior camera calibrations.
Problem with prior camera calibration based solutions is that, once the relative position of
projector-camera or display surface changes, camera calibration becomes void and recalibration
has to be performed. This makes them unsuitable for dynamic scenarios.
In this work, feasibility study is performed and two techniques for geometric correction of 2D
quadric surfaces are proposed. One technique is hybrid whereas other is passive online. Proposed
hybrid technique requires offline display area estimation phase whereas passive online technique
does not require prior camera calibration. In both techniques, moving least squares deformation
is used to calculate the Bezier feature points. Based upon these feature points next images are
prewarped using Bezier transformation. Moreover a line based metric to estimate geometric
correction achieved is also introduced. After successful simulation, the algorithm is tested on real
setup. It is also shown that, the work can be extended to dynamic surfaces.
