US 12,364,905 B2
Synthesis of exercise guidance training data
Giuseppe Barbalinardo, Berkeley, CA (US); Joshua Ben Shapiro, Toronto (CA); Asim Kadav, Mountain View, CA (US); Ivan Savytskyi, Mississauga (CA); Rajiv Bhan, Mountain View, CA (US); Rustam Paringer, Samara (RU); and Aly E. Orady, Austin, TX (US)
Assigned to Tonal Systems, Inc., San Francisco, CA (US)
Filed by Tonal Systems, Inc., San Francisco, CA (US)
Filed on Oct. 16, 2023, as Appl. No. 18/380,585.
Claims priority of provisional application 63/417,052, filed on Oct. 18, 2022.
Prior Publication US 2024/0123284 A1, Apr. 18, 2024
Int. Cl. A63B 24/00 (2006.01); G06T 7/73 (2017.01); G06V 20/40 (2022.01); G06V 40/20 (2022.01)
CPC A63B 24/0062 (2013.01) [A63B 24/0075 (2013.01); G06T 7/73 (2017.01); G06V 20/41 (2022.01); G06V 40/23 (2022.01); A63B 2024/0065 (2013.01); G06T 2200/24 (2013.01); G06T 2207/10016 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/30196 (2013.01)] 24 Claims
OG exemplary drawing
 
1. A system, comprising:
a communication interface;
a camera;
a processor coupled to the communication interface and the camera, and wherein the processor is configured to:
receive from the communication interface a first video of a first individual performing an exercise movement using a first camera viewpoint, wherein the first video is associated with a first guidance label;
receive from the camera a second video of a second individual with a second camera viewpoint;
wherein the first video is in a video domain and the second video is in the video domain;
generate a modified video of the second individual with the second camera viewpoint at least in part by passing the first video to a pose data change model at least in part by:
extracting a motion of the first individual from the first video using key point extraction in the pose domain; and
motion retargeting the extracted motion of the first individual to the second individual in the second video with the second camera viewpoint at least part by mapping motion between different kinematic structures;
wherein the modified video generated is in the video domain; and
wherein the pose data change model is configured to perform global data changes or local data changes;
associate the modified video with a second guidance label to match the first guidance label; and
train a guidance classifier model using the modified video.