Method, apparatus and computer program for generating sports game highlight video based on winning probability

A method for generating a sports game highlight video based on winning probability, comprising the steps of:

• receiving video of the sports game from at least one image capture device;
• performing object tracking analysis on the video to generate tracking data providing ball trajectory and player position information;
• identifying a plurality of shot events in the video of the sports game based on said tracking data, wherein each shot event represents a different stage of a shot, and the shot events correspond to one or more of a tennis ball being hit by a first player, the tennis ball passing a net, the tennis ball landing on a court, and the tennis ball being hit by a second player;
• extracting video segments of the identified shot events from the video of the sports game;
• generating shot event metadata for indexing the video segments;
• computing a winning probability matrix comprising winning probabilities at point level by a winning probability model of the sports game, based on said shot event metadata;
• computing a point excitement matrix based on said shot event metadata, wherein the point excitement matrix comprises one or more excitement weights pertaining to one or more of shot variety, final shot quality, rally length, overall shot quality, winners and errors, return improbability, or average unreturnability occurring within a rally;
• and
• generating a sports game highlight video based on said winning probability matrix and point excitement matrix,
• wherein the step of generating a sports game highlight video further includes:
• ranking video segments according to deviations of actual outcomes of the sports game from the corresponding winning probabilities in said winning probability matrix,
• concatenating the ranked video segments,
• generating tracking model videos from the tracking data so that the tracking model video includes visual presentations of the tracking data, and
• merging video segments with tracking model videos corresponding to the same shot.