Incremental focus of attention for robust visual tracking

We present the Incremental Focus of Attention (IFA) architecture for adding robustness to software-based, real-time, motion trackers. The framework provides a structure which, when given the entire camera image to search, efficiently focuses the attention of the system into a narrow set of possible slates that includes the target state. IFA offers a means for automatic tracking initialization and reinitialization when environmental conditions momentarily deteriorate and cause the system to lose track of its target. Systems based on the framework degrade gracefully as various assumptions about the environment are violated. In particular, multiple tracking algorithms are layered so that the failure of a single algorithm causes another algorithm of less precision to take over, thereby allowing the system to return approximate feature state information.