Structure and motion from uniform 3D acceleration

The author derives a multiframe feature-based closed form solution to the structure from motion problem when the camera and an object (or the environment) are undergoing uniform relative 3D acceleration. A central projection camera model is assumed, and no smoothness assumptions are imposed concerning object surfaces. It is assumed that a separate process has already performed motion correspondence and segmentation, and that sequences of features across at least three images are provided as input. The algorithm can process not only point features, but also line features as well. In addition, a given feature is not required to be visible in every frame. The (serial) computational complexity of the algorithm is proportional to the number of features being tracked times the number of frames. Test results are presented for synthetic and real data