Touchless wrist measurement
Various implementations disclosed herein include devices, systems, and methods that determine a wrist measurement or watch band size using depth data captured by a depth sensor from one or more rotational orientations of the wrist. In some implementations, depth data captured by a depth sensor including at least two depth map images of a wrist from different angles is obtained. In some implementations, an output is generated based on inputting the depth data into a machine learning model, the output corresponding to circumference of the wrist or a watch band size of the wrist. Then, a watch band size recommendation is provided based on the output.
1. A method comprising:
at an electronic device having a processor:
obtaining depth data captured by a depth sensor, the depth data comprising a set of depth map images of a wrist from different angles, wherein the set of depth map images are captured with different portions of the wrist facing the depth sensor;
selecting at least two depth map images from the set based on determining that the two depth map images correspond to at least a minimum threshold difference in viewpoint with respect to an angle of the different angles;
generating an output based on inputting the depth data into a machine learning model, the output corresponding to circumference of the wrist or a watch band size of the wrist; and
providing a watch band size recommendation based on the output.
2. The method of claim 1 , wherein the machine learning model is a convolutional neural network regressor.
3. The method of claim 1 , wherein a segmentation mask identifying portions of the depth data corresponding to the wrist is input to the machine learning model, wherein the segmentation mask is generated based on a light-intensity sensor data.
4. The method of claim 1 , wherein the machine learning model is trained using real and synthetic training data.
5. The method of claim 1 , wherein the machine learning model is trained using training data that identifies a wrist circumference at a plurality of arm locations.
6. The method of claim 1 , wherein the machine learning model is trained using training data that corresponds to watch tightness.
7. The method of claim 1 , wherein the machine learning model is trained based on weighted forearm measurements, wherein weights of the weighted forearm measurements correspond to relative significance of the weighted forearm measurements along an axis of an arm.
8. The method of claim 1 , wherein the machine learning model further outputs a confidence value corresponding to a confidence in the circumference of the wrist or the watch band size of the wrist.
9. The method of claim 1 , wherein the depth data comprises a two-view depth pair of two depth map images.
10. The method of claim 9 , wherein the depth data comprises additional depth map images before, between, and after the two depth map images.
11. The method of claim 1 , wherein the at least two depth map images are captured as the wrist is rotated in front of the depth sensor.
12. The method of claim 1 , wherein guidance regarding positioning of the wrist or depth sensor is provided while the depth data is obtained.
13. The method of claim 1 , wherein the at least two depth map images each comprise depth values for portions of a hand.
14. The method of claim 1 , wherein the electronic device is a mobile phone or tablet.
15. The method of claim 1 , wherein the watch band size recommendation is associated with a non-adjustable watch band.
16. The method of claim 1 , wherein the at least a minimum threshold difference in viewpoint is selected based on identifying at least two different ellipse parameters representing shape and size of the wrist to determine the circumference of the wrist.
17. The method of claim 1 , wherein the at least two depth map images further comprise portions of an arm located above the wrist.
18. A system comprising:
memory; and
one or more processors at a device coupled to the memory, wherein the memory comprises program instructions that, when executed on the one or more processors, cause the system to perform operations comprising:
obtaining depth data captured by a depth sensor, the depth data comprising a set of depth map images of a wrist from different angles, wherein the set of depth map images are captured with different portions of the wrist facing the depth sensor;
selecting at least two depth map images from the set based on determining that the two depth map images correspond to at least a minimum threshold difference in viewpoint with respect to an angle of the different angles;
generating an output based on inputting the depth data into a machine learning model, the output corresponding to circumference of the wrist or a watch band size of the wrist; and
providing a watch band size recommendation based on the output.
19. The system of claim 18 , wherein the at least two depth map images are selected based on determining that the at least two depth map images correspond to at least a threshold difference in rotation of the wrist around a longitudinal axis of an arm.
20. A non-transitory computer-readable storage medium, storing program instructions executable via one or more processors to perform operations comprising:
obtaining depth data captured by a depth sensor, the depth data comprising a set of depth map images of a wrist from different angles, wherein the set of depth map images are captured with different portions of the wrist facing the depth sensor;
selecting at least two depth map images from the set based on determining that the two depth map images correspond to at least a minimum threshold difference in viewpoint with respect to an angle of the different angles;
generating an output based on inputting the depth data into a machine learning model, the output corresponding to circumference of the wrist or a watch band size of the wrist; and
providing a watch band size recommendation based on the output.