Method for efficiently tracking object models in video sequences via dynamic ordering of features

摘要: An object model has a plurality of features and is described by state. unregistered feature the model, an available frame from sequence images are selected to minimize cost function subsequent search for match image in frame. Upon match, registered that The state then updated each steps selecting, searching updating repeated. A video storage module may contain only one corresponding single time instance, which case framework used based on integrated sequential selection. Alternatively, store entire sequence, selection performed across all frames maximum tracking efficiency. Finally, small number previous plus current frame, spans subset matching cycle.