In order to obtain the application of three dimensional
(3D) reconstruction techniques for disaster visualization system
on the mountain. This research is to develop a 3D model of
mountain from triple images, and try to develop real time
reconstruction process of 3D-mountain. The processes of this
research study consist of seven stages to develop the 3D model’s
mountain surface. In feature points detection, by using Scaleinvariant
feature transform (SIFT), the matched points all of
rescale and full scale images are the same number of matched
points. Compare the result of accuracy of correspondent matched
points, the SITF appropriate more than Harris corner detector.
The detected feature points of Fuji’s mountain are covered the
entire surface but it could not be detected on the edge of
mountain. Because of using feature point extraction on far
distance image (about 10 km), matching results were not clear,
and then the 3D model package was used to help in register
image as well as help to find the relative rotation and translation
of both images. The fundamental metric is derived from the
Maximum a posteriori sample consensus (MAPSAC)
implementation that is a new method for the robust estimations
with absolutely no additional computational burden. The result
of this study can be made 3D model of Fuji’s mountain the
accuracy about 10 m.