Key point recognition with uncertainty measurement

A method, comprising:

• obtaining an image that depicts an object having a key point;
• determining, based on the image, an estimated key point location for the key point;
• determining, based on the image, an uncertainty value that represents accuracy of the estimated key point location;
• and
• determining a filtered key point location estimate by updating a previous key point location estimate based on the estimated key point location and the uncertainty value;
• and
• controlling operation of an automated system based in part on the filtered key point location estimate.

A method, comprising:

• determining, based on an image, an estimated key point location for a key point of an object;
• determining, based on the image, an uncertainty value that represents accuracy of the estimated key point location, wherein the uncertainty value includes a first standard deviation value for the first spatial direction of the image and a second standard deviation value for the second spatial direction of the image;
• and
• controlling operation of an automated system based in part on the estimated key point location and the uncertainty value.

A system, comprising:

• a sensor that is configured to obtain a series of images that depict an object having a key point;
• an image analyzer that is configured to determine, for image from the series of images, an estimated key point location for the key point, and an uncertainty value that represents accuracy of the estimated key point location;
• and
• an object tracker that is configured to determine a filtered key point location estimate by updating a previous key point location estimate based on the estimated key point location and the uncertainty value and determine a position estimate for the object and an orientation estimate for the object based on the filtered key point location estimate.