Machine learning augmented loop drive training

Patent metadata

Field	Value
Publication number	US-11369844-B2
Application number	US-202017038535-A
Country	US
Kind code	B2
Filing date	Sep 30, 2020
Priority date	Sep 30, 2020
Publication date	Jun 28, 2022
Grant date	Jun 28, 2022

Abstract

Disclosed are techniques for leveraging machine learning to generate posture adjustment values for specific body postures of a player to improve loop drive techniques, such as in table tennis. Video clips of a player hitting a ball with a loop drive technique are analyzed to determine values for specific body postures and qualities of the ball after being hit. A machine learning model is generated to analyze relationships between body posture values and ball qualities. Upon receiving a video clip of a live session of a player hitting a ball using a loop drive technique, the machine learning model is used to generate adjustment values for body postures of the player to impart improved loop drive qualities to the ball, such as faster topspin.

First claim

What is claimed is:

1. A computer-implemented method (CIM) comprising:
receiving a plurality of historical loop drive session data sets, where a loop drive session data set includes at least a player posture data set and a corresponding ball return quality data set, where a given player posture data set includes a plurality of posture data points, and a given ball return quality data set includes a plurality of ball return quality metrics;
generating a table tennis training machine learning model based, at least in part, on the plurality of loop drive session data sets;
receiving a live loop drive session data set corresponding to a loop drive session of a live table tennis player; and
assigning labels to the plurality of posture data points of the player posture data set of the live loop drive session data set based, at least in part, on the table tennis training machine learning model and a ball return quality data set of the live loop drive session data set.

2. The CIM of claim 1 , further comprising:
flagging at least one posture data point of the plurality of posture data points of the player posture data set of the live loop drive session data set for posture adjustment based, at least in part, on the assigned label corresponding to the at least one posture data point.

3. The CIM of claim 2 , further comprising:
determining a quantitative posture adjustment value for the flagged at least one posture data point based, at least in part, on the table tennis training machine learning model and the ball return quality data set of the live loop drive session data set.

4. The CIM of claim 3 , further comprising:
outputting the determined quantitative posture adjustment value to the live table tennis player.

5. The CIM of claim 4 , wherein:
the outputted determined quantitative posture adjustment value is outputted audibly using text-to-speech techniques through a speaker device; and
the outputted determined quantitative posture adjustment value is outputted visually on an electronic display device.

6. The CIM of claim 1 , wherein:
the plurality of posture data points includes two data subsets corresponding to: (i) a pre-contact posture data set corresponding to table tennis player posture data points prior to striking a table tennis ball with a table tennis racket, and (ii) a contact posture data set corresponding to table tennis player posture data points at or after striking the table tennis ball with the table tennis racket;
the pre-contact posture data set includes: (i) an angle value corresponding to a direction and degree of turning in a waist of a table tennis player, (ii) a height distance value corresponding to a distance between shoulders of the table tennis player and a floor surface the table tennis player is standing on, (iii) an angle value corresponding to a direction and degree of bending at knees of the table tennis player between their upper and lower legs, and (iv) a velocity value corresponding to a table tennis racket speed;
the contact posture data set includes: (i) an angle value corresponding to a direction and degree of bending at an elbow of the table tennis player in an arm holding the table tennis racket, (ii) an angle value corresponding to a direction and degree of bending at a shoulder of the table tennis player between their upper arm of the arm holding the table tennis racket and their torso, (iii) a height distance value corresponding to height of an elbow of the arm of the table tennis player holding the table tennis racket from a table tennis table, (iv) a set of racket posture values corresponding to height of the table tennis racket from the table tennis table, degree and direction of an angle between a striking surface of the table tennis racket and the table tennis ball, and position of the table tennis ball on the table tennis racket striking surface when striking occurs between the table tennis racket and table tennis ball, and (v) velocity and acceleration values for the table tennis racket when striking occurs and through a follow-through motion; and
the plurality of ball return quality metrics includes: (i) a top/downspin rotation speed value of the table tennis ball after striking occurs, (ii) a ball velocity value after contacting the table tennis table after striking occurs, (iii) a side spin rotation speed value of the table tennis ball after striking occurs, (iv) a score value corresponding to whether the table tennis scored from striking the table tennis ball.

7. A computer program product (CPP) comprising:
a machine readable storage device; and
computer code stored on the machine readable storage device, with the computer code including instructions for causing a processor(s) set to perform operations including the following:
receiving a plurality of historical loop drive session data sets, where a loop drive session data set includes at least a player posture data set and a corresponding ball return quality data set, a given player posture data set includes a plurality of posture data points, and a given ball return quality data set includes a plurality of ball return quality metrics,
generating a table tennis training machine learning model based, at least in part, on the plurality of loop drive session data sets,
receiving a live loop drive session data set corresponding to a loop drive session of a live table tennis player, and
assigning labels to the plurality of posture data points of the player posture data set of the live loop drive session data set based, at least in part, on the table tennis training machine learning model and a ball return quality data set of the live loop drive session data set.

8. The CPP of claim 7 , wherein the computer code further includes instructions for causing the processor(s) set to perform the following operations:
flagging at least one posture data point of the plurality of posture data points of the player posture data set of the live loop drive session data set for posture adjustment based, at least in part, on the assigned label corresponding to the at least one posture data point.

9. The CPP of claim 8 , wherein the computer code further includes instructions for causing the processor(s) set to perform the following operations:
determining a quantitative posture adjustment value for the flagged at least one posture data point based, at least in part, on the table tennis training machine learning model and the ball return quality data set of the live loop drive session data set.

10. The CPP of claim 9 , wherein the computer code further includes instructions for causing the processor(s) set to perform the following operations:
outputting the determined quantitative posture adjustment value to the live table tennis player.

11. The CPP of claim 10 , wherein:
the outputted determined quantitative posture adjustment value is outputted audibly using text-to-speech techniques through a speaker device; and
the outputted determined quantitative posture adjustment value is outputted visually on an electronic display device.

12. The CPP of claim 7 , wherein:
the plurality of posture data points includes two data subsets corresponding to: (i) a pre-contact posture data set corresponding to table tennis player posture data points prior to striking a table tennis ball with a table tennis racket, and (ii) a contact posture data set corresponding to table tennis player posture data points at or after striking the table tennis ball with the table tennis racket;
the pre-contact posture data set includes: (i) an angle value corresponding to a direction and degree of turning in a waist of a table tennis player, (ii) a height distance value corresponding to a distance betwe

Assignees

• IBM

Inventors

• Ran Chuan
• Wang Junhui
• Zhu Guandong
• Cui Yongchang
• Lu Shuyan

Classifications

• A63B2208/0204

  Standing on the feet · CPC title

• A63B2024/0009

  Computerised real time comparison with previous movements or motion sequences of the user · CPC title

• A63B2102/16

  Table tennis · CPC title

• A63B2024/0034

  during flight · CPC title

• A63B2220/05

  Image processing for measuring physical parameters · CPC title