×

System and Method for Identifying, Analyzing, and Reporting on Players in a Game from Video

  • US 20170154222A1
  • Filed: 11/25/2016
  • Published: 06/01/2017
  • Est. Priority Date: 11/26/2015
  • Status: Active Grant
First Claim
Patent Images

1. A method for improved automated processing of a single image frame or multiple image frames acquired from a video source, said processing being performed by one or more computers and comprising the following computer-implemented steps of:

  • a) selecting an image frame from said single image frame or from said multiple image frames,b) creating multispectral ratios, multispectral indices, and multispectral transformations from the wavebands of said selected image frame from step (a),c) producing a field of play mask using polychotomous classification of said multispectral ratios, multispectral indices, and multispectral transformations from step (b),d) identifying field of play marking object candidates from said selected image frame from step (a) using polychotomous classification of said multispectral ratios, multispectral indices, and multispectral transformations from step (b) under said field of play mask from step (c),e) creating univariate object shape measures of said field of play marking object candidates from step (d),f) creating field of play markings with topology from said field of play marking object candidates from step (d) using conditional expressions applied to said multispectral ratios, multispectral indices, and multispectral transformations from step (b) and said univariate object shape measures from step (e), or ANN modelling inputs of said multispectral ratios, multispectral indices, and multispectral transformations from step (b) and said univariate object shape measures from step (e), or the Radon transform,g) creating xy image control points from erosion of said field of play markings with topology from step (f),h) creating field of play control lines with topology using linear regression modelling from said field of play markings with topology from step (f),i) creating xy image control points with topology from intersections of said field of play control lines with topology from step (h),j) building a point to point file by combining said image control points from steps (g) and (i) and pairing said image control points from steps (g) and (i) with the predefined field of play reference control points,k) creating a perspective projection image to field of play georeferencing transformation model from said point to point file from step (j),l) identifying player object candidates from said selected image frame using polychotomous classification of said multispectral ratios, multispectral indices, and multispectral transformations from step (b) under said field of play mask from step (c),m) creating univariate object shape measures of the said player object candidates from step (l),n) outputting and labelling player object fragments by team from said player object candidates from step (l) using ANN models of inputs containing said multispectral ratios, multispectral indices, and multispectral transformations from step (b) and said univariate object shape measures from step (m),o) creating XY Cartesian player fragment locations by team, registered to the field of play, by applying the said transformation model from step (k) to said player object fragments by team from step (n),p) aggregate said XY Cartesian player fragment locations by team from step (o) to identify single XY Cartesian player locations for each player on each team, andq) save player locations, team identifier, the perspective projection to planimetric transformation model, control points, rink topology, and image frame identifier on a non-transitory computer readable medium.

View all claims
  • 0 Assignments
Timeline View
Assignment View
    ×
    ×