System and method for human hand motion detection by skin color prediction
First Claim
1. A method of detecting whether a selected pixel in an electronic image F(t), having a pixel color, represents human skin in motion, comprising:
- calculating a first probability, P1, indicating a likelihood that the color of the selected pixel appears within a first distribution of pixel colors observed in human skin, wherein the color of the selected pixel and the first distribution of pixel colors are represented in the CbCr color space;
calculating a second probability, P2, indicating a likelihood that the selected pixel color appears within a second distribution of pixel colors that are observed outside of human skin, wherein the second distribution of pixel colors are represented in the CbCr color space; and
calculating a third probability, PK, indicating a probability that the selected pixel represents human skin, wherein PK is a function of P1 and P2;
wherein the selected pixel is determined to represent human skin in motion if an intensity difference, I, between the intensities of the selected pixel within F(t) and a temporally adjacent electronic image F(t−
Δ
t) is greater than a threshold intensity value, T, which is a function of PK.
1 Assignment
0 Petitions
Accused Products
Abstract
Systems and methods for motion detection of human skin within temporally adjacent electronic images are provided. Motion detection is accomplished by analyzing the color values of selected pixels within the electronic images represented in CbCr color space. Histogram distributions which represent skin colors and non-skin colors in CbCr color space are modeled in order to provide likelihoods that a selected color value appears within the histogram distributions. Posterior skin probability values, indicating the probability that a selected pixel having a given CbCr color value represents human skin, are calculated from these likelihoods. For each of the selected pixels, an intensity difference of the pixel between the electronic images is compared to an adaptive intensity threshold which is a function of the posterior skin probability in order to determine whether the pixel is in motion.
6 Citations
25 Claims
-
1. A method of detecting whether a selected pixel in an electronic image F(t), having a pixel color, represents human skin in motion, comprising:
-
calculating a first probability, P1, indicating a likelihood that the color of the selected pixel appears within a first distribution of pixel colors observed in human skin, wherein the color of the selected pixel and the first distribution of pixel colors are represented in the CbCr color space; calculating a second probability, P2, indicating a likelihood that the selected pixel color appears within a second distribution of pixel colors that are observed outside of human skin, wherein the second distribution of pixel colors are represented in the CbCr color space; and calculating a third probability, PK, indicating a probability that the selected pixel represents human skin, wherein PK is a function of P1 and P2; wherein the selected pixel is determined to represent human skin in motion if an intensity difference, I, between the intensities of the selected pixel within F(t) and a temporally adjacent electronic image F(t−
Δ
t) is greater than a threshold intensity value, T, which is a function of PK. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9)
-
-
10. A system for identifying the motion of a selected pixel having a pixel color within a plurality of temporally adjacent electronic images, comprising:
-
a skin likelihood calculator, wherein the skin likelihood calculator employs a skin color model which models a first histogram distribution of pixel colors representing human skin in order to calculate a skin likelihood, P1, representing a likelihood of observing the color of the selected pixel in human skin; and wherein the skin likelihood calculator further employs a non-skin color model which models a second histogram distribution of pixel colors which do not represent human skin in order to calculate a non-skin likelihood, P2, representing a likelihood of observing the color of the selected pixel outside of human skin wherein the pixel color of the selected pixel and the first and second histogram distributions are represented in the CbCr color space; a posterior probability calculator which employs a Bayesian equation, the color of the selected pixel, and the skin and non-skin likelihoods P1 and P2 in order to calculate a probability that the selected pixel represents human skin; and a motion detection engine which determines that the selected pixel represents human skin in motion if a difference in intensity of the selected pixel between a first electronic image and a second electronic image selected from the plurality of electronic images is greater than a threshold intensity value, T, wherein T is a function of PK. - View Dependent Claims (11, 12, 13, 14, 15)
-
-
16. A method of detecting the motion of human skin within a video, comprising:
-
receiving at least a first video frame and a second video frame which are temporally adjacent to one another; selecting a pixel from the first video frame, the pixel having a pixel color; calculating a skin probability, PK that the selected pixel represents human skin, wherein the skin probability PK is a function of a likelihood of observing the selected pixel'"'"'s color in a first plurality of pixels representing human skin and a likelihood of observing the selected pixel'"'"'s color in a second plurality of pixels not representing human skin, wherein the pixel color of the selected pixel and the first and second plurality of pixels are represented in the CbCr color space; comparing an intensity difference, I, between the selected pixel within the first and second video frames to a threshold intensity, T, comprising a function of PK; wherein the selected pixel is considered to represent the motion of human skin if the value of I is greater than T. - View Dependent Claims (17, 18, 19, 20, 21, 22)
-
-
23. A system for detecting human motion within a video, comprising:
-
a video source capable of transmitting a plurality of video frames; and a motion evaluation system which receives at least a first and a second temporally adjacent video frames from the video source, the motion evaluation system comprising a skin probability calculator and a motion detection engine; wherein the skin probability calculator calculates a posterior skin probability, PK, for at least one selected pixel within the plurality of video frames represents human skin, the posterior skin probability PK comprising a probability of observing the pixel color of the selected pixel in a first histogram distribution of pixel colors representing human skin, PS, and a probability of observing the pixel color of the selected pixel in a second histogram distribution of pixel colors which do not represent human skin, PNS, wherein the pixel color of the selected pixel and the first and second histogram distributions are represented in the CbCr color space; and wherein motion detection engine determines that the selected pixel is in motion if an intensity difference of the selected pixel between the first electronic image and the second electronic image is greater than a threshold intensity, T, which comprises a function of PK. - View Dependent Claims (24, 25)
-
Specification