Model-based multi-hypothesis target tracker
AssigneesIntel Corporation, Omek Interactive, Ltd.
InventorsShahar Fleischmann, Yotam Livny, Gideon Kleimer, Gershom Kutliroff
Life of this patent
Estimating the term…
Abstract
The present disclosure describes a target tracker that evaluates frames of data of one or more targets, such as a body part, body, and/or object, acquired by a depth camera. Positions of the joints of the target(s) in the previous frame and the data from a current frame are used to determine the positions of the joints of the target(s) in the current frame. To perform this task, the tracker proposes several hypotheses and then evaluates the data to validate the respective hypotheses. The hypothesis that best fits the data generated by the depth camera is selected, and the joints of the target(s) are mapped accordingly.
Claims26
A method for tracking a target in a sequence of depth images, the method comprising:
- acquiring with a depth sensor the sequence of depth images;
- formulating multiple hypotheses for a target configuration in a selected depth image of the sequence of depth images, wherein each of the multiple hypotheses includes a set of corresponding points between a known target configuration in a previous depth image in the sequence of depth images and the target configuration in the selected depth image;
- using a target skeleton model, synthesizing a depth map of the target configuration in the selected depth image for each of the multiple hypotheses;
- testing each hypothesis by comparing the respective synthesized depth map to the selected depth image to identify a best hypothesis that most closely fits the target configuration in the selected depth image.
A method for tracking a user's hand in a sequence of depth images, the method comprising:
- acquiring with a depth sensor the sequence of depth images, wherein the sequence of depth images includes a preceding depth image and current depth image;
- calibrating a hand skeleton model to the hand;
- segmenting the hand in the current depth image from a rest of the depth image;
- determining a current hand pose for the current depth image based on a known pose of the hand in the preceding depth image, the calibrated hand skeleton model, and the segmented hand in the current depth image.
A system of tracking a hand in a sequence of depth images, the system comprising:
- a depth sensing module configured to acquire a sequence of depth images of a hand;
- a tracking module configured to track movements of the hand in the sequence of depth images, wherein tracking movements comprises:
- identifying multiple hypotheses for hand configurations in a current depth image;
- rendering a skeleton model of the hand for each of the multiple hypotheses;
- applying an objective function used to identify a best hypothesis that most closely fits the hand configuration in the current depth image.
A system for tracking a target in a sequence of depth images, the method comprising:
- means for acquiring the sequence of depth images;
- a tracking module configured to track movements of the target in the sequence of depth images, wherein tracking movements comprises:
- formulating multiple hypotheses for a target configuration in a selected depth image of the sequence of depth images, wherein each of the multiple hypotheses includes a set of corresponding points between a known target configuration in a previous depth image in the sequence of depth images and the target configuration in the selected depth image;
- synthesizing a depth map of the target configuration in the selected depth image for each of the multiple hypotheses;
- testing each hypothesis by comparing the respective synthesized depth map to the selected depth image to identify a best hypothesis that most closely fits the target configuration in the selected depth image.
Dependencies are inferred from the claim text.