System and Method for Identifying, Analyzing, and Reporting on Players in a Game from Video
First Claim
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.
0 Assignments
0 Petitions
Accused Products
Abstract
The invention is a system and method used for detection, analysis, and reporting on player metrics of a sporting event using video images. The main components of the invention are: grabbing a digital video image frame or grabbing an analog video image frame and converting the analog video image frame to a digital video image frame, extracting field of play markings from the video image frame(s) for use as image control points, creating a perspective projection registration model based upon the pairing of image control points with a set of user-defined control points representative of the field of play, extracting player image locations on each team from the video image frames, and applying the perspective projection transformation model to register participant image locations to their respective planimetric coordinate position on the user defined field of play. In the case of processing multiple video image frames, to correct errors across perspective projection transformation models developed on each image frame, a final bundle adjustment affine transformation model is applied to correct participant locations. Analysis of players use network analysis, pattern analysis, spatial interpolation, hypothesis testing, or forecast modeling can then be performed with reports in tabular, chart and graphic, or cartographic formats.
-
Citations
19 Claims
-
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, and q) 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 Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16)
-
-
17. A method of producing an automatic final bundle adjustment model to correct participant location errors across multiple image frames that have been analyzed to detect and locate players from opposing teams among a plurality of sport participants in said multiple image frames, the method comprising the following computer-implemented steps of:
-
a) transforming image control points from each of said multiple image frames using a respective image frame georeferencing transformation model and relating said transformed image control points to all predefined field of play reference control points, b) building a final bundle adjustment model using results from step (a), c) applying said final bundle adjustment model from step (b) on single Cartesian XY player locations for each player on each team, and d) storing said final bundle adjustment model from step (b) and final adjusted player location results from step (c) in non-transitory computer readable memory. - View Dependent Claims (18, 19)
-
Specification