Analysis of Sporting Events Using Softbots

A supervised softbot utilized for analyzing, segmenting, and properly classifying video clips pertaining to a wide variety of sporting events is presented. First, selected action scenes (i.e., training sequences) of a given sporting event are automatically segmented into real-world objects representing the participants of the activity. These objects correspond to the players, the playing field (or court) and required equipment used within the event. The most active regions are manually selected and represent the training objects. Color and texture features are then extracted from each of the training objects tracked across several frames of the training sequence... An algorithm that matches features between objects in the training video and objects in all other sequences (within the same video) is performed and subsequently classifies the sequence as a candidate sporting sequence. Motion data is integrated into classification algorithm and provides a key feature necessary for identifying the degree of activity of the sporting clip. Only sequences that exhibit the same degree of motion as the training sequence will be classified as an active sports scene. The softbot system is tested on a wide range of sporting events with results provided.