Feature Flow Fields in Out-of-Core Settings

Feature Flow Fields (FFF) are an approach to tracking features in a time-dependent vector field v. The main idea is to introduce an appropriate vector field f in space-time, such that a feature tracking in v corresponds to a stream line integration in f . The original approach of feature tracking using FFF requested that the complete vector field v is kept in main memory. Especially for 3D vector fields this may be a serious restriction, since the size of time-dependent vector fields can exceed the main memory of even high-end workstations. We present a modification of the FFF-based tracking approach which works in an out-of-core manner. For an important subclass of all possible FFF-based tracking algorithms we ensure to analyze the data in one sweep while holding only two consecutive time steps in main memory at once. Similar to the original approach, the new modification guarantees the complete feature skeleton to be found. We apply the approach to tracking of critical points in 2D and 3D time-dependent vector fields.