Composing FlyCam Panoramic Images

How does it work? Just smashing together video images from two adjacent cameras results in obvious seams. To get a convincing panorama, we need to correct for many things. We correct for lens distortion by unwarping each image using a piecewise-bilinear or radial transformation as shown: 

In operation, "patches" from the raw camera inputs (left) are unwarped into undistorted squares. Patches on camera borders are then cross-faded to yield the composite panorama (above). All operations are done in real-time on multiple video streams.

Another method to improve the image is to correct for color and brightness differences between cameras. Because our inexpensive cameras have automatic gain control, images of the same area can differ in brightness. The simplest fix for this is to cross-fade images, as shown above, however we have developed more sophisticated techniques for color and brightness correction. The image shown below contains these and other enhancements.

Here's a paper describing  more technical details of the FlyCam system. Jonathan Foote and Don Kimber: "FlyCam: Practical Panoramic Video," in Proc. IEEE International Conference on Multimedia and Expo, August 2000.

Return to the FlyCam page