Tilt variance method of detecting and accounting for true motion of imagery subject to atmospheric turbulence

A method of determining a probability a pixel in a video image subject to optical turbulence represents true motion, the method comprising the steps of:

• a. capturing incoming light from a scene with an image capture device to yield an original video having a sequence of frames;
• b. creating a prototype image from the sequence of frames, the prototype image being a geometrically accurate representation of the scene and substantially free of contamination caused by true scene motion;
• c. measuring a local atmospheric turbulence profile from local statistical data;
• d. selecting a subject pixel location from each pixel location available in the prototype image;
• e. determining a turbulence tilt variance for the subject pixel location based upon the measured local atmospheric turbulence profile;
• f. extracting a finite window for each subject pixel location based on the turbulence tilt variance, whereby the subject pixel location is contained within the window;
• g. establishing a unique mode for each pixel in the finite window whereby the intensity of each pixel is a mean of the respective mode and a weight for each unique mode is based upon the turbulence tilt variance to yield a Gaussian mixture model;
• h. evaluating each pixel at the subject pixel location in the original image sequence against the probability density function from the Gaussian mixture model to determine a probability that each pixel at the subject pixel location represents true motion within the original video;
• repeating steps e through h until all pixels locations within the prototype image have been evaluated and probabilities therefor have been determined and further comprising the step of converting the probability density function for each pixel, to a negative log likelihood value proportional to the likelihood that the subject pixel represents true motion within the original image.

A method of determining a probability a pixel in a video image subject to optical turbulence represents true motion, the method comprising the steps of:

• a. capturing incoming light from a scene with an image capture device to yield an original video having a sequence of frames;
• b. creating a prototype image from the sequence of frames, the prototype image being a geometrically accurate representation of the scene and substantially free of contamination caused by true scene motion;
• c. estimating a local atmospheric turbulence profile;
• d. selecting a subject pixel location from each pixel location available in the prototype image;
• e. determining a turbulence tilt variance for the subject pixel location based upon the measured local atmospheric turbulence profile;
• f. extracting a finite window for each subject pixel location based on the turbulence tilt variance, whereby the subject pixel location is contained within the window;
• g. establishing a unique mode for each pixel in the finite window whereby the intensity of each pixel is a mean of the respective mode and a weight for each unique mode is based upon the turbulence tilt variance to yield a Gaussian mixture model;
• h. evaluating each pixel at the subject pixel location a subject original frame against the probability density function from the Gaussian mixture model to determine a probability that the pixel at the subject pixel location in the sequence of frames represents true motion within the original video;
• i. repeating step h until all pixels at that subject pixel location have been evaluated and probabilities therefor have been determined;
• j. selecting a different subject pixel location;
• and k. repeating steps e through j in sequence until all pixels within the sequence of frames have been evaluated and probabilities for each pixel location of the video image have been determined.

A method for determining the likelihood a pixel in a video image subject to optical turbulence represents true motion, the method comprising the steps of:

• capturing incoming light from a scene with an image capture device to yield an original image having a sequence of frames;
• creating a prototype image from the original image as a geometrically accurate representation of the scene, the prototype image being substantially free of contamination caused by true scene motion;
• creating a tilt variance-Gaussian mixture model based upon optical turbulence statistics associated with the scene;
• deriving a likelihood function from the tilt variance-Gaussian mixture model;
• and
• evaluating each pixel in the original image sequence against the prototype image using the likelihood function to yield a likelihood that the respective pixel represents true motion.

A method for determining a likelihood a pixel in a video image subject to optical turbulence represents true motion, the method comprising the steps of:

• capturing incoming light from a scene with an image capture device to yield an original image having a sequence of frames;
• creating a prototype image from the original image as a geometrically accurate representation of the scene, the prototype image being substantially free of contamination caused by true scene motion;
• creating a tilt variance-Gaussian mixture model based upon optical turbulence statistics associated with the scene;
• deriving a likelihood function from the tilt variance-Gaussian mixture model;
• and evaluating each pixel in the original image sequence against the prototype image using the likelihood function to yield a likelihood that the respective pixel represents true motion, wherein the likelihood is negative log likelihood.