Abstract: This paper presented a method to robustly and rapidly reconstruct general 3D scene from 2D images in camera, with the appropriate virtual application to show the effects of reconstruction. We presented novel camera calibralion and scene reconstruction using scale-invariant feature points. Scale invariant feature has impressive matching ability in camera's vidco,which greatly reduces the restriction of facility in camera calibration, broadens the area of video reconstruction. With some optimization, we enhanced the accuracy of reconstruction and speed. At last, we processed the visualized application, which shows the effects of our system. The result illustrates that our system is capable of producing accurate and fast scene structure from video in a very few seconds.

Key words: Feature matching, Camera calibration, 3D reconstruction, Virtual reality, Structure from motion