Panoramas of the images taken on Kaidan head came out very nicely. The RANSAC algorithm aligns the pairs of images very well. The blending by a 2D hat function also yielded satisfactory results. Adjusting the power of the hat function to generate steeper transition made the panorama a little clearer.
When images contained humans moving around, it seemed the alignment was sometimes disturbed by them. I tried raising the power of the weights to make the transition in blending steeper. This fixed the ghosted humans, but then the steeper illumination changes made the panorama less attractive.
The hand held sequences tend to generate less accurate matches, due to much greater perturbations in the process of taking pictures.