Structure Based Feature Extraction in Basketball Zone-Defense Strategies

This paper proposes a framework for structure-based feature extraction in basketball zonedefense strategies. Firstly, a graphical representation for key-frames extracted from zone-defense video clips is introduced, where each key-frame is expressed in terms of a zone-defense graph, representing the positions of defenders and the ball. Secondly, defense-lines are defined and extracted from zone-defense graphs for each zone-defense strategy, based on which, a 10dimentional feature vector with respect to the defense-lines is introduced to characterize the structure relationships. Experiments have been conducted for basketball zone-defense strategy detection on both simulated and real-life basketball zone-defense video database, which demonstrate the validation and practicability of such a structure based feature characterization, and, in particular, its robustness with respect to the disturbance of local transformation of subprime nodes in the graphs.