CPC G06F 3/017 (2013.01) [G06T 7/13 (2017.01); G06T 7/60 (2013.01); G06V 10/443 (2022.01); G06V 10/46 (2022.01); G06V 10/50 (2022.01); G06V 40/113 (2022.01); G06V 40/20 (2022.01)] | 26 Claims |
1. A method for recognizing a gesture in real-time, the method being performed in a device having access to a database of histograms, each histogram being a shape descriptor of a shape contour of a gesture image, each shape descriptor and histogram containing weights of the contour pixels, the weights being calculated by means of the contour pixels within Angular Radial Bins, ARBs, the device is communicating with a sensor adapted to capture an image of a gesture, the method comprising the steps of:
determining a contour pixels image from the captured image of a query gesture;
calculating an ARB distribution adapted to the contour pixels image of the gesture;
determining a histogram by calculating a shape descriptor for each of an angle tilted instance (n) of the ARB distribution belonging to a sequence of angle tilted instances (n=0, 1, 2, . . . , N−1) of the ARB distribution, wherein the weight hn(i,j) of each ARB (i,j) is based on the contour pixels within the ARB(i,j);
comparing the determined histogram of the captured gesture image to the histograms of the database; and
determining a gesture by selecting the histogram of the database best matching the calculated histogram.
|