A Novel Method for Geometric Correction of Multi-cameras in Panoramic Video System

This paper describes a novel method based on self-adaptive subdivision mesh for geometry correction of multi-cameras in a panoramic video system. The system hardware consists of an array of six board cameras pointing outwardly. Digitally combining synchronized frames from each camera, a wide-field panoramic video can be created. Using sparse initial corresponding points between the ideal target surface and camera image, the proposed method can subdivide the grid into a dense one in arbitrary precision. Hence, the bijection between camera image and projection surface can be established in pixel level. This mapping can be used to warp a camera image to ideal panorama surface. The method is robust to lens distortion and camera relative position, and can seamlessly integrate with image matting processing. With the GPU processing, the real time frame rate panoramic video can be available. The experiment proves the effectiveness and simpleness of the method.