Transforming sports implement motion sensor data to two-dimensional image for analysis
3D motion sensor data from a sensor on an athletic implement such as a golf club is clipped around an impact event and the clip then reduced to its 2D components, which are combined into a single 2D image and provided to a machine learning algorithm to output an indication of a flaw in the motion (e.g., a flaw in a golf swing).
1. An apparatus comprising:
at least one processor configured to access at least one computer storage with instructions executable by the processor to:
receive from at least one motion sensor three-dimensional (3D) motion data;
identify a continuous sequence in the motion data from a time before a time of impact to a time after the time of impact;
identify a trajectory during the continuous sequence of an implement with which the motion sensor is engaged;
transform the trajectory into x-y, y-z, and z-x planes to render three position images;
determine velocity components in the x-y, y-z, and z-x planes transformed from the trajectory to render three orientation images;
combine the orientation images and position images into a single two dimensional (2D) planar image;
input the planar image to at least one machine learning (ML) engine; and
receive as output from the ML engine in response to inputting the planar image an indication of at least one flaw in moving the implement.
2. The apparatus of claim 1 , wherein the instructions are executable to determine the velocity components using swing decomposition from address to impact.
3. The apparatus of claim 1 , comprising the motion sensor.
4. The apparatus of claim 1 , wherein the implement comprises a golf club.
5. The apparatus of claim 1 , wherein the implement comprises a tennis racket.
6. The apparatus of claim 1 , wherein the implement comprises a table tennis paddle.
7. The apparatus of claim 1 , wherein the implement comprises a baseball bat.
8. The apparatus of claim 1 , wherein the implement comprises a hockey stick.
9. A method comprising:
receiving information from a motion sensor representing motion of an implement in three dimensions (3D) in a time domain;
transforming the information in the time domain to information in a 2D image domain at least in part by identifying a trajectory using the information in the time domain, transforming the trajectory into x-y, y-z, and z-x planes to render three position images, determining velocity components in the x-y, y-z, and z-x planes to render three orientation images, and combining the orientation images and position images to establish the information in the 2D image domain; and
using the information in the 2D image domain to output a characterization of motion of the implement.
10. The method of claim 9 , wherein the characterization of motion comprises identifying a flaw from a set of “N” flaws, wherein “N” is an integer.
11. The method of claim 9 , comprising:
identifying using signals from the motion sensor, a time of impact;
identifying a continuous sequence in the information from the motion sensor from a time before the time of impact to a time after the time of impact; and
identifying a trajectory during the continuous sequence of an implement with which the motion sensor is engaged.
12. The method of claim 9 , comprising:
inputting the information in the 2D image domain to at least one machine learning (ML) engine; and
receiving from the ML engine the characterization of motion of the implement.
13. The method of claim 12 , wherein the ML engine comprises at least one image classification model.
14. The method of claim 9 , comprising:
inputting the information in the 2D image domain to at least one machine learning (ML) engine; and
receiving as output from the ML engine an indication of at least one flaw in moving the implement.