Tracking interacting hands using sensor of wearable multimedia device
View Patent ↗Systems, methods, devices and non-transitory, computer-readable storage mediums are disclosed for the tracking of interacting hands using a sensor of a wearable multimedia device. In an embodiment, a method comprises: obtaining, with a sensor of a wearable multimedia device, a first frame of two-dimensional (2D) image data and a second frame of three-dimensional (3D) depth data; determining multiple hand regions in the first frame of 2D image data; determining a location of each hand region in the first frame of 2D image data; detecting at least one landmark in each detected hand region; generating a confidence score for each landmark detected in each detected hand region; determining 3D world coordinates for each landmark that is visible in the first frame of the 2D image data; tracking each landmark in 3D world coordinates; and determining an interaction with the hands based on the tracking of each landmark in 3D world coordinates.
1 . A method comprising:
obtaining, with a sensor of a wearable multimedia device, a first frame of two-dimensional (2D) image data and a second frame of three-dimensional (3D) depth data;
determining, with at least one processor, multiple hand regions in the first frame of 2D image data;
determining, with the at least one processor, a location of each hand region in the first frame of 2D image data;
detecting, with the at least one processor, at least one landmark in each detected hand region;
generating, with the at least one processor, a confidence score for each landmark detected in each detected hand region;
determining, with the at least one processor, 3D world coordinates for each landmark that is visible in the first frame of the 2D image data;
tracking, with the at least one processor, each landmark in 3D world coordinates; and
determining, with the at least one processor, an interaction with the hands based on the tracking of each landmark in 3D world coordinates.
2 . The method of claim 1 , wherein the at least one landmark is a joint or finger tip of a finger in the detected hand region.
3 . The method of claim 1 , wherein the at least one landmark is a palm or wrist in the detected hand region.
4 . The method of claim 1 , wherein detecting at least one landmark in each detected hand region comprises using a landmark detection model that is trained to predict the 2D or 3D point coordinates of the at least one landmark.
5 . The method of claim 1 , wherein bounding boxes for the location of each hand region in the first frame of 2D image data and for the at least one landmark in each detected hand region are predicted by a machine learning model, and the bounding boxes identify locations of the hand regions and landmarks in the 2D image data.
6 . The method of claim 5 , wherein the landmark detection model is trained on annotated ground truth data and a synthetic hand model over various backgrounds that is mapped to corresponding 2D or 3D point coordinates.
7 . The method of claim 1 , wherein the sensor is a time of flight camera that outputs infrared or amplitude image data and the depth data that are registered to each other by the sensor.
8 . The method of claim 7 , wherein determining 3D world coordinates in the second frame of 3D depth data for each landmark that is visible in the first frame of the 2D image data comprises adding a depth component from the depth data to the corresponding 2D pixel coordinates of each landmark.
9 . The method of claim 7 , wherein the time of flight camera is an infrared camera that is adjusted to measure a range of temperature that approximates human body temperature, and each hand region is detected by binarization on the first frame using a threshold value.
10 . The method of claim 1 , wherein each hand region is detected using template matching.
11 . The method of claim 1 , wherein the landmark is determined to be visible based at least in part on the confidence score for the landmark.
12 . A system comprising:
a sensor;
at least one processor; and
memory storing instructions that, when executed by the at least one processor, cause the at least one processor to perform operations comprising:
obtaining, with the sensor, a first frame of two-dimensional (2D) image data and a second frame of three-dimensional (3D) depth data;
determining multiple hand regions in the first frame of 2D image data;
determining a location of each hand region in the first frame of 2D image data;
detecting at least one landmark in each detected hand region;
generating a confidence score for each landmark detected in each detected hand region;
determining 3D world coordinates for each landmark that is visible in the first frame of the 2D image data;
tracking each hand in 3D world coordinates; and
determining an interaction with the hands based on the tracking of each landmark in 3D world coordinates.
13 . The system of claim 12 , wherein the at least one landmark is a joint or finger tip of a finger in the detected hand region.
14 . The system of claim 12 , wherein the at least one landmark is a palm in the detected hand region.
15 . The system of claim 12 , wherein detecting at least one landmark in each detected hand region comprises using a landmark detection model that is trained to predict the 2D or 3D world coordinates of the at least one landmark.
16 . The system of claim 12 , wherein bounding boxes for the location of each hand region in the first frame of 2D image data and for the at least one landmark in each detected hand region are predicted by a machine learning model, and the bounding boxes identify locations of the hand regions and landmarks in the 2D image data.
17 . The system of claim 16 , wherein the landmark detection model is trained on annotated ground truth data and a synthetic hand model over various backgrounds that is mapped to corresponding 3D world coordinates.
18 . The system of claim 17 , wherein determining 3D world coordinates in the second frame of 3D depth data for each landmark that is visible in the first frame of the 2D image data comprises adding a depth component from the depth data to the corresponding 2D pixel coordinates of each landmark.
19 . The system of claim 12 , wherein the sensor is a time of flight camera that outputs infrared or amplitude image data and the depth data that are registered by the sensor.
20 . The system of claim 19 , wherein the time of flight camera is an infrared camera that is adjusted to measure a range of temperature that approximates human body temperature, and each hand region is detected by binarization on the first frame using a threshold value.
21 . The system of claim 12 , wherein each hand region is detected using template matching.
22 . The system of claim 12 , wherein the landmark is determined to be visible based at least in part on the confidence score for the landmark.