作者: M. Han , T. Kanade
关键词: Motion estimation 、 Match moving 、 Iterative reconstruction 、 Artificial intelligence 、 Mathematics 、 Motion field 、 Structure from motion 、 Computer vision 、 Camera auto-calibration 、 Representation (mathematics) 、 Real image
摘要: We describe a reconstruction method of multiple motion scenes, which are the scenes containing moving objects, from uncalibrated views. Assuming that objects with constant velocities, recovers scene structure, trajectories camera and intrinsic parameters (except skews) simultaneously. The number is automatically detected without prior segmentation. based on unified geometrical representation static objects. It first performs projective using bilinear factorization algorithm, then converts solution to Euclidean one by enforcing metric constraints. Experimental results synthetic real images presented.