Hand detection and tracking method and device
For each frame of a video, a determination is made whether an image of a hand exists in the frame. When at least one frame of the video includes the image of the hand, locations of the hand in the frames of the video are tracked to obtain a tracking result. A verification is performed to determine whether the tracking result is valid in a current frame of the frames of the video. When the tracking result is valid in the current frame of the video, a location of the hand is tracked in a next frame. When the tracking result is not valid in the current frame, localized hand image detection is performed on the current frame.
1. A computer-implemented method, comprising:
determining, for each frame of a video, whether an image of a hand exists in the frame;
when at least one frame of the video includes the image of the hand, tracking locations of the hand in the frames of the video to obtain a tracking result;
verifying that the tracking result is not valid in a current frame of the frames of the video, and
performing localized hand image detection on the current frame by:
determining a center of a positioning block used to identify a location of the image of the hand in the frame,
defining a plurality of adjacent blocks by using a specified step size and a specified block size,
adjusting each of the plurality of adjacent blocks to a size determined during a hand training, and
providing, to an image object classifier, each adjusted adjacent block to determine a quantity of the plurality of adjacent blocks that include images of hands.
2. The computer-implemented method of claim 1 , wherein determining whether the image of the hand exists in the frame comprises:
scanning the frame to detect a presence of the image of the hand to produce scans; and
performing, using the scans, hand image detection by using a histogram of oriented gradients (HOG)+support vector machines (SVM) method at different scales.
3. The computer-implemented method of claim 1 , wherein tracking the locations of the hand in the frames of the video to obtain the tracking result comprises using a template matching method to obtain the tracking result, the template matching method based on a template determined from a location of the image of the hand in the frame.
4. The computer-implemented method of claim 1 , wherein, when the quantity of the plurality of adjacent blocks that include images of hands is two or more, using each of the adjacent blocks for subsequent tracking.
5. The computer-implemented method of claim 1 , wherein, when the quantity of the plurality of adjacent blocks that include images of hands is one, using a combination of the adjacent block with the positioning block for subsequent tracking.
6. The computer-implemented method of claim 1 , wherein, when the quantity of the plurality of adjacent blocks that include images of hands is zero, performing additional hand image detection.
7. A non-transitory, computer-readable medium storing one or more instructions executable by a computer system to perform operations comprising:
determining, for each frame of a video, whether an image of a hand exists in the frame;
when at least one frame of the video includes the image of the hand, tracking locations of the hand in the frames of the video to obtain a tracking result;
verifying that the tracking result is not valid in a current frame of the frames of the video, and
performing localized hand image detection on the current frame by:
determining a center of a positioning block used to identify a location of the image of the hand in the frame,
defining a plurality of adjacent blocks by using a specified step size and a specified block size,
adjusting each of the plurality of adjacent blocks to a size determined during a hand training, and
providing, to an image object classifier, each adjusted adjacent block to determine a quantity of the plurality of adjacent blocks that include images of hands.
8. The non-transitory, computer-readable medium of claim 7 , wherein determining whether the image of the hand exists in the frame comprises:
scanning the frame to detect a presence of the image of the hand to produce scans; and
performing, using the scans, hand image detection by using a histogram of oriented gradients (HOG)+support vector machines (SVM) method at different scales.
9. The non-transitory, computer-readable medium of claim 7 , wherein tracking the locations of the hand in the frames of the video to obtain the tracking result comprises using a template matching method to obtain the tracking result, the template matching method based on a template determined from a location of the image of the hand in the frame.
10. The non-transitory, computer-readable medium of claim 7 , wherein, when the quantity of the plurality of adjacent blocks that include images of hands is two or more, using each of the adjacent blocks for subsequent tracking.
11. The non-transitory, computer-readable medium of claim 7 , wherein, when the quantity of the plurality of adjacent blocks that include images of hands is one, using a combination of the adjacent block with the positioning block for subsequent tracking.
12. The non-transitory, computer-readable medium of claim 7 , wherein, when the quantity of the plurality of adjacent blocks that include images of hands is zero, performing additional hand image detection.
13. A computer-implemented system, comprising:
one or more computers; and
one or more computer memory devices interoperably coupled with the one or more computers and having tangible, non-transitory, machine-readable media storing one or more instructions that, when executed by the one or more computers, perform one or more operations comprising:
determining, for each frame of a video, whether an image of a hand exists in the frame;
when at least one frame of the video includes the image of the hand, tracking locations of the hand in the frames of the video to obtain a tracking result;
verifying that the tracking result is not valid in a current frame of the frames of the video, and
when the tracking result is valid in the current frame of the video, tracking a location of the hand in a next frame; and
when the tracking result is not valid in the current frame, performing localized hand image detection on the current frame by:
determining a center of a positioning block used to identify a location of the image of the hand in the frame,
defining a plurality of adjacent blocks by using a specified step size and a specified block size,
adjusting each of the plurality of adjacent blocks to a size determined during a hand training, and
providing, to an image object classifier, each adjusted adjacent block to determine a quantity of the plurality of adjacent blocks that include images of hands.
14. The computer-implemented system of claim 13 , wherein determining whether the image of the hand exists in the frame comprises:
scanning the frame to detect a presence of the image of the hand to produce scans; and
performing, using the scans, hand image detection by using a histogram of oriented gradients (HOG)+support vector machines (SVM) method at different scales.
15. The computer-implemented system of claim 13 , wherein tracking the locations of the hand in the frames of the video to obtain the tracking result comprises operations of using a template matching method to obtain the tracking result, the template matching method based on a template determined from a location of the image of the hand in the frame.
16. The computer-implemented system of claim 13 , wherein:
when a quantity of the plurality of adjacent blocks that include images of hands is two or more, using each of the adjacent blocks for subsequent tracking;
when a quantity of the plurality of adjacent blocks that include images of hands is one, using a combination of the adjacent block with the positioning block for subsequent tracking; and
when a quantity of the plurality of adjacent blocks that include images of hands is zero, performing additional hand image detection.