Extending knowledge data in machine vision

Abstract

A machine-vision system configured to detect a first feature associated with a virtual-sporting event, detect a second feature associated with the virtual-sporting event, provide the first detected feature and the second detected feature as input to a machine-vision system to identify at least a first portion of a representation of the virtual-sporting event, and combine the first detected feature and the second detected feature to validate the first portion of the representation of the virtual-sporting event.

1 . A computer-implemented method comprising:
 detecting a first feature of a first dataset associated with a virtual-sporting event as a first detected feature;   detecting a second feature of the first dataset associated with the virtual-sporting event as a second detected feature;   providing the first detected feature and the second detected feature as input to a machine-vision system to identify at least a first portion of a representation of the virtual-sporting event; and   combining the first detected feature and the second detected feature to validate the first portion of the representation of the virtual-sporting event.   
     
     
         2 . The computer-implemented method of  claim 1 , wherein:
 the first detected feature includes a center of a representation of a player;   the second detected feature includes a silhouette of the representation of the player; and   the method further comprising generating a bounding box feature associated with the representation of a player of the virtual-sporting event.   
     
     
         3 . The computer-implemented method of  claim 1 , wherein the first detected feature and the second detected feature include environment features, the method further comprising:
 predicting a third environment feature based on the first detected feature and the second detected feature;   determining a location and an orientation to present the third environment feature as part of a determined presentation; and   identifying whether presentation of the third environment feature corresponds with other parts of the determined presentation.   
     
     
         4 . The computer-implemented method of  claim 1 , further comprising:
 detecting a time feature associated with the virtual-sporting event as a detected time feature; and   identifying an error in a second portion of the representation of the virtual-sporting event based on the detected time feature and validation of the first portion of the representation of the virtual-sporting event.   
     
     
         5 . The method of  claim 1 , wherein:
 a first machine-learning component detects the first feature;   a second machine-learning component detects the second feature; and   the first machine-learning component and the second machine-learning component are different types of machine-learning components.   
     
     
         6 . The method of  claim 5 , wherein as different types of machine-learning components associated with different feature types increase, reliability of the machine-vision system improves. 
     
     
         7 . The method of  claim 1 , wherein at least part of at least one of the first dataset or a second dataset associated with an actual-sporting event includes differential image data. 
     
     
         8 . The computer-implemented method of  claim 1 , further comprising:
 selecting a portion of recognized results from a second dataset for annotation;   annotating the portion of the recognized results from the second dataset to generate an annotated portion;   adding the annotated portion to the first dataset to obtain an extended dataset;   determining whether learned data from the extended dataset represents improved detection of features compared to the first dataset;   providing the extended dataset to the machine-vision system;   identifying an error in the learned data based on the first detected feature being outside of a bounding box;   sorting the extended dataset according to a number of errors in the learned data to obtain a sorted dataset; and   providing a middle portion of the sorted dataset as a trained-synthetic dataset.   
     
     
         9 . The method of  claim 8 , wherein the second dataset is associated with an actual-sporting event and includes video data. 
     
     
         10 . A computer-implemented method comprising:
 detecting a first feature of a first dataset associated with a representation of a virtual-sporting event as a first-detected feature;   detecting a second feature of the first dataset associated with the representation of the virtual-sporting event as a second-detected feature;   generating a bounding box associated with the first-detected feature and the second-detected feature;   selecting a portion of recognized results from a second dataset for annotation;   annotating the portion of the recognized results from the second dataset to generate an annotated portion;   adding the annotated portion to the first dataset to obtain an extended dataset;   determining that learned data from the extended dataset represents improved detection of features compared to the first dataset;   providing the extended dataset to a machine-vision system;   identifying an error in the learned data based on the first-detected feature being outside of the bounding box;   sorting the extended dataset according to a number of errors in the learned data to obtain a sorted dataset; and   providing a middle portion of the sorted dataset as a trained-synthetic dataset.   
     
     
         11 . The computer-implemented method of  claim 10 , wherein:
 the first-detected feature includes a center of a representation of a player from the representation of the virtual-sporting event;   the second-detected feature includes a silhouette of the representation of the player from the representation of the virtual-sporting event; and   the bounding box is associated with the representation of the player from the representation of the virtual-sporting event.   
     
     
         12 . The computer-implemented method of  claim 10 , wherein at least one of the first-detected feature and the second-detected feature include an environment feature, the method further comprising:
 predicting a third-environment feature based on the first-detected feature and the second-detected feature;   determining a location and an orientation to present the third-environment feature as part of a determined presentation; and   identifying whether presentation of the third-environment feature corresponds with other parts of the determined presentation.   
     
     
         13 . The computer-implemented method of  claim 12 , further comprising:
 detecting a time feature associated with the first dataset as a detected-time feature; and   identifying an error in the representation of the virtual-sporting event based on the detected-time feature and the trained-synthetic dataset.   
     
     
         14 . The computer-implemented method of  claim 10 , wherein:
 a first machine-learning component detects the first feature;   a second machine-learning component detects the second feature; and   the first machine-learning component and the second machine-learning component are different types of machine-learning components.   
     
     
         15 . The computer-implemented method of  claim 14 , wherein as different types of machine-learning components associated with different feature types increase, reliability of the machine-vision system improves. 
     
     
         16 . The computer-implemented method of  claim 10 , wherein at least part of at least one of the first dataset or the second dataset includes differential image data. 
     
     
         17 . A system comprising:
 one or more processors; and   one or more computer-readable media storing computer-executable instructions that, when executed by the one or more processors, configure a machine-vision system to perform operations to:
 detect a first feature associated with a representation of a virtual-sporting event as a first-detected feature of a first dataset; 
 detect a second feature associated with the representation of the virtual-sporting event as a second-detected feature of the first dataset; 
 generate a bounding box associated with the first-detected feature and the second-detected feature; 
 select a portion of recognized results from a second dataset for annotation; 
 annotate the portion of the recognized results from the second dataset to generate an annotated portion; 
 add the annotated portion to the first dataset to obtain an extended dataset; 
 determine that learned data from the extended dataset represents improved detection of features compared to the first dataset; 
 provide the extended dataset to the machine-vision system; 
 identify an error in the learned data based on the first-detected feature being outside of the bounding box; 
 sort the extended dataset according to a number of errors in the learned data to obtain a sorted dataset; and 
 provide a middle portion of the sorted dataset as a trained-synthetic dataset. 
   
     
     
         18 . The system of  claim 17 , wherein the first dataset includes a time feature; and
 the machine-vision system is configured to identify an error in a first portion of the representation of the virtual-sporting event based on the time feature and validation of a second portion of the representation of the virtual-sporting event.   
     
     
         19 . The system of  claim 17 , wherein a first machine-learning component detects the first feature and a second machine-learning component detects the second feature. 
     
     
         20 . The system of  claim 17 , wherein:
 the first dataset includes a tagged part of a representation of a player; and   the tagged part includes at least one of a head of the representation of the player, a foot of the representation of the player, or a hockey stick associated with the representation of the player.