3D Reconstruction from Image Sequence Based on Scene Merger Method

We have established a practical 3D reconstruction system from image sequence. For the first step, spare 3D points are reconstructed from the first two images according to stereo vision theory. This reconstruction result is called a scene. Similarly, reconstruction from the every two adjacent images is achieved. However, each scene depends on its own coordinate system and they are not coinciding. In order to relate these reconstruction results, we proposed a scene merger strategy by estimating the scale factor, rotation matrix and translation vector between two scenes; and finally get image sequences' reconstruction.