Motion intensity and direction distribution similarity measure for accelerating sports video retrieval

To accelerate sports video retrieval, an efficient approach of describing motion intensity is necessary. However, the existing methods suffer from high complexity for relating the motion features to goal events. This short paper proposes a novel video description and similarity measure for sports video retrieval. Sports video retrieval is accomplished based on the motion direction descriptor (MDD) derived from variance-based classification of motion intensity and direction. Similarity is computed by the proposed parametric distance evaluation. The contribution of this work is its low computation and adaption of motion direction to goal event detection. Based on query-by-example experiments, the proposed video retrieval approach has higher performance when compared with that of other algorithms.