Video Synthesis with multiple cameras
A wide area virtual city model with scalability can be built by individual restored multiple three-dimensional structures and camera paths by extended factorization-method to all round picture sequences which is one of the techniques of Structure Form-method, and by applying bundle adjustment anew. ( Fig.1)
In the automatic grouping of an affiliated sequence picture, measure motions each video after eliminating blank points after RANSAC and DP matching is performed by using space- time EPI/THI. ( Fig.2)
Fig.3 shows the result of view of the building from the ground, which is configure solid plane and given texture mapping. The point groups in the air are feature point and they are acquiring colors from original pictures that are pursued. The red line aroundthe map is a movement of locus by cameras from vehicles and it is estimated at the same time.
Fig. 1: Factorization_method Fig. 2: Viewpoint from city construction result and the ground (West Shinjuku) Fig. 3 Building the solid plane, the point groups which exist in .

Publications
    [1] Ryota Matsuhisa , Shintaro Ono,Hiroshi Kawasaki,Atsuhiko Banno,Katsushi Ikeuchi "Structure from Motion for omnidirectional images using efficient factorization method based on virtual camera rotation" International Workshop Computer Vision and Its Application to Image Media Processing, (CD-ROM), pp.1-7, 1.2009

    [2] Ryota Matsuhisa , Hiroshi Kawasaki,Shintaro Ono,Atsuhiko Banno,Katsushi Ikeuchi
    "Extensive urban city model construction using multiple omnidirectional image sequences taken by vehicle camera (in Japanese)"
    Meeting on Image Recognition and Understanding 2009 (MIRU 2009), pp.1570-1577, 7.2008 (Best Interactive Session Paper Award)
Kawasaki Laboratory