Non-central Catadioptric Camera Calibration Using Generalized Unified Model

Omni-directional catadioptric cameras featuring the advantage of large field of view over the traditional perspective cameras are being increasingly used in lots of fields such as robotics navigation, surveillance, video Conferencing, etc. Depending on whether they pose a single viewpoint, catadioptric cameras can be classified as central or non-central imaging systems. Central catadioptric cameras are ideal but not realistic. Non-proper relative position between the mirror and the camera will cause misalignment, which makes cental catadioptric camera non-central. Existing models for central catadioptric cameras, such as unified projection model, degenerate quickly while misalignment becomes large. And existing models for non-central catadioptric cameras are always very complex, and hard to calibrate or use. Therefore, it’s urgent to find a novel model which keeps the simplicity but works well even under severe misalignment.In this paper, a generalized unified model(GUM) has been present. By adding two parameters to the unified projection model, the GUM can compensate the misalignment quite well. It works well even under severe misalignment, while maintaining the simplicity of unified projection model. This makes it easy to calibrate and use non-central catadioptric camera.Based on the GUM, we give out a method for calibration with planar grids. The model and method works well even under severe misalignment. We test our model with both synthetic images and real images. Both experiments prove the validity of our model and calibration method.With the GUM’s forward projection, backward projection and epipolar geometry, a full self-calibration method has been proposed. The GUM can be calibrated with two images automatically. Also, simulation and real experiment verify the effectiveness of GUM and the self-calibration method.