Self-calibration from multiple views with a rotating camera
ECCV '94 Proceedings of the third European conference on Computer vision (vol. 1)
True Multi-Image Alignment and Its Application to Mosaicing and Lens Distortion Correction
IEEE Transactions on Pattern Analysis and Machine Intelligence
Characterization of errors in compositing panoramic images
Computer Vision and Image Understanding
International Journal of Computer Vision
A Flexible New Technique for Camera Calibration
IEEE Transactions on Pattern Analysis and Machine Intelligence
Automated Mosaics via Topology Inference
IEEE Computer Graphics and Applications
Gradient-based approach for fine registration of panorama images
Journal of Computer Science and Technology - Special issue on computer graphics and computer-aided design
Motion Panoramas: Research Articles
Computer Animation and Virtual Worlds
| Hi-index | 0.00 |
An approach is presented that can simultaneously correct focal length error and improve pair-wise registration for the task of panorama composition from a sequence of photographs taken from a fixed view position by a rotating camera. In building such panorama image covering a full view, a 3-parameter rotation model with a roughly estimated focal length is frequently used to relate adjacent images in the sequence. However, the pair-wise registration thus obtained may not be perfect due to the error of focal length. Whereas, accumulated errors in subsequent pairs may result in an obvious end-gap or overlap when wrapping up the images onto the viewing surface. By analyzing the relationship between focal length and panning angle parameter of the pair-wise registration model, we propose a new method that can iteratively refine both the focal length as well as the panning angles at the same time. The approach produces physically correct pair-wise registration and seamlessly closed end-gap in final panorama composition. The advantages of the method are (a) no reliance on prior camera calibration; (b) no repeat of pair-wise registration and panorama construction with the updated focal length each round, which considerably reduces the computational cost. Experiments show that our method can obtain the same visual effect as re-performing image registration, but much faster. Results of the algorithm are presented.