Target acquisition and tracking system

Target detection method comprising the steps of:

• determining joint edge feature data and for mapping the same into a probability of target with confidence bounds based upon a maximum likelihood statistic experimentally derived from a training set of known examples of known imagery,
• determining target-to-interference ratio matched filter feature data in such a manner so to be independent or anticorrelated to said MLC means, and for mapping the same into a probability of target with confidence bounds based a maximum likelihood statistic experimentally derived from said training set,
• determining grey level feature data in such a manner as to be independent or anticorrelated singly and jointly to said MLC and GTIR steps, and for mapping said grey level feature data into a probability of target with confidence bounds based upon a maximum likelihood statistic experimentally derived from said training set, and
• combining or mapping the target detection outputs from said MLC, GTIR, and VSC steps into a composite probability of target with confidence bounds based upon a maximum likelihood statistic experimentally derived from said training set.

Target detection apparatus comprising

• first means for determining first feature data and for mapping the same into a probability of target with confidence bounds based upon a maximum likelihood statistic experimentally derived from a training set of known examples of known imagery to give a first list of targets,
• second means for determining second ratio matched feature data in such a manner so to be independent or anticorrelated to said first means, and for mapping the same into a probability of target with confidence bounds based a maximum likelihood statistic experimentally derived from said training set to give a second list of targets,
• said first and second feature data being statistically independent or anticorrelated,
• means for combining or mapping the target lists from said first and second means into a composite probability of target with confidence bounds based upon a maximum likelihood statistic experimentally derived from said training set, and
• target prioritization means for ranking the target output of said means to yield a target ranking list.

In a target detection method for detecting targets in a sensed image to determine the correlation probability of objects in the image to stored feature recognition data, comprising

• generating a feature pattern corresponding to a first algorithmic point of view of the scene,
• matching the feature pattern with a stored set of feature recognition data of the same algorithmic point of view to determine the correlation probability therewith,
• generating one or more other feature patterns from one or more other points of view of the scene which are independent of the first point of view in all combinations, both singly and jointly,
• matching the other feature patterns with other stored sets of feature recognition data of the other points of view,
• said points of view being represented by algorithms different from each other in such ways as to generate correlated true detection but independent of anticorrelated false detections in all combinations, both singly and jointly.

Apparatus for detecting targets in a sensed image to determine the correlation probability of objects in the image to stored feature recognition data, comprising

• means for generating a feature pattern corresponding to a first algorithmic point of view of the scene,
• means for matching the feature pattern with a stored set of feature recognition data of the same algorithmic point of view to determine the correlation probability therewith,
• means for generating one or more other feature patterns from one or more other points of view of the scene which are independent of the first point of view in all combinations, both singly and jointly,
• means for matching the other feature patterns with other stored sets of feature recognition data of the other points of view,
• said points of view being represented by algorithms different from each other in such ways as to generate correlated true detection but independent of anticorrelated false detections in all combinations, both singly and jointly.

A target detection method comprising

• determining joint edge events and for mapping the same into a measure of probability of target based on a maximum likelihood junction derived from a training set,
• determining target-to-interference ratio data and for mapping the same into a measure of probability of target based on a matched filter function,
• determining spatial clustering features and for mapping the same into a measure of probability of target based on a geometric function,
• said MLC, GTIR, VSC, and HPL steps being capable of being trained by a set of known targets in sets of exemplar visual fields to develop a knowledge base,
• said MLC, GTIR, and VSC steps being substantially singly and jointly independent of or anticorrelating of each other in all combinations,
• HPL combining the target detection outputs from said MLC, GTIR, and VSC means into a mapping function based upon the training data,
• so that said system is ready to process and identify unknown targets in new visual fields,
• running a training set including known targets each of the above process steps to develop a knowledge base of extrapolated known situations subsequently processing new visual fields with unknown targets using said knowledge base through each of said MLC, GTIR, VSC, and HPL steps.

A target detection system comprising

• MLC means for determining joint edge events and for mapping the same into a measure of probability of target based on a maximum likelihood function derived from a training set to develop a knowledge base,
• GTIR means for determining target-to-interference ratio data and for mapping the same into a measure of probability of target based on a matched filter function derived from a training set to develop a knowledge base,
• VSC means for determining spatial clustering features and for mapping the same into a measure of probability of target based on a geometric function derived from a training set to develop a knowledge base,
• said MLC, GTIR, and VSC means being substantially singly and jointly independent of or anticorrelating of each other in all combinations,
• HPL means for combining the target detection outputs from said MLC, GTIR, and VSC means into a mapping function based upon the training data, said MLC, GTIR, VSC, and HPL means being capable of being trained by a set of known targets in sets of visual fields to develop a knowledge base,
• so that said system is ready to process and identify unknown targets in new visual fields.

A target detection apparatus comprising

• GTIR means for determining target-to-interference ratio matched filter feature data in such a manner so to be independent or anticorrelated to other target detection means, and for mapping the same into a probability of target with confidence bounds based on a maximum likelihood statistic experimentally derived from said training set to select a target,
• said GTIR means including
• means for determining the average local contrast difference between target related picture elements and background related picture elements,
• means for determining the local grey level variation, and
• means for generating a target to interference related function from said local contrast difference and said local grey level variations.

A target detection apparatus comprising

• MLC means for determining joint feature data and for mapping the same into a probability of target with confidence bounds based upon a maximum likelihood statistic experimentally derived from a training, of known examples of known imagery to select a target,
• GTIR means for determining target-to-interference ratio matched filter feature data in such a manner so to be independent or anticorrelated to other target detection means, and for mapping the same into a probability of target with confidence bounds based on a maximum likelihood statistic experimentally derived from said training set to select a target,
• VSC means for determining grey level feature data in such a manner as to be independent or anticorrelated singly and jointly to said MLC and GTIR means, and for mapping said grey level feature data into a probability of target with confidence bounds based upon a maximum likelihood statistic experimentally derived from said training set to select a target,
• HPL means for combining (or mapping) said target selections from said means into a composite probability of target with confidence bounds based upon a maximum likelihood statistic experimentally derived from a training set,
• means for determining if an input data cluster contains a probability of target assessment,
• means for determining the types of probability assessments,
• means for forwarding MLC probability of target data to a one dimensional decision Parzen table,
• means for forwarding VSC data to a one dimensional decision Parzen table,
• means for merging MLC probability target and VSC probability of target and data,
• a two dimensional Parzen table responsive to said merged MLC and VSC data,
• means for merging MLC probability of target and GTIR probability of target and data,
• a two feature Parzen table responsive to said merged MLC and GTIR data,
• a single feature Parzen table responsive to GTIR probability of target data,
• means to merge GTIR probability of target and VSC probability of target and data,
• a two feature Parzen table responsive to said merged GTIR and VSC probability of target data,
• means to merge MLC, GTIR, and VSC probability of target, and
• a three feature Parzen table responsive to said merged MLC, GTIR, and VSC probability of target data, and
• target prioritization means for ranking the target output of said HPL means to yield a target ranking list.

A target detection method comprising

• a GTIR step for determining target-to-interference ratio matched filter feature data in such a manner so to be independent or anticorrelated to said MLC means, and for mapping the same into a probability of target with confidence bounds based on a maximum likelihood statistic experimentally derived from said training set to select a target,
• determining the average local contrast difference between target related picture elements and background related picture elements,
• determining the local grey level variation, and
• generating a target to interference related function from said local contrast difference and said local grey level variations.

A target detection method comprising

• a VSC step for determining grey level feature data, and for mapping said grey level feature data into a probability of target with confidence bounds based upon a maximum likelihood statistic experimentally derived from said training set to select a target, including the steps of
• developing a grey level histogram of a target sized window,
• calculating upper and lower threshold levels for said histogram as a function of histogram characteristics,
• eliminating picture elements falling below the low level value,
• calculating the centroid of a cluster of pixels exceeding said threshold value,
• creating a global histogram from the entire video image,
• calculating a first optimal global histogram from said global histogram and said upper grey level threshold,
• calculating a second global histogram function from said lower grey level threshold,
• calculating a target sized histogram from said target sized histogram data and said low level threshold value, and
• a look-up table step responsive to said upper threshold level global histogram said lower threshold level global histogram said target sized window histogram and said centroid for producing a probability of target function and a confidence function.

Method for detecting targets in a sensed image to determine the correlation probability of objects in the image to stored feature recognition data, comprising

• generating a feature pattern corresponding to a first algorithmic point of view of the scene,
• matching the feature pattern with a stored set of feature recognition data of the same algorithmic point of view to determine the correlation probability therewith,
• generating one or more other feature patterns from one or more other points of view of the scene which are independent of the first point of view in all combinations, both singly and jointly,
• matching the other feature patterns with other stored sets of feature recognition data of the other points of view, said points of view being represented by algorithms different from each other in such ways as to generate correlated true detection but independent of anticorrelated false detections in all combinations, both singly and jointly.

A target detection apparatus comprising

• VSC means for determining grey level feature data and for mapping grey level feature data into a probability of target with confidence bounds based upon a maximum likelihood statistic experimentally derived from said training set to select a target,
• said VSC means including
• means for developing a grey level histogram of a target sized window,
• means for calculating upper and lower threshold levels for said histogram as a function of histogram characteristics,
• means for eliminating picture elements falling below the low level value,
• means for calculating the centroid of a cluster of pixels exceeding said threshold value,
• means for creating a global histogram from the entire video image,
• means for calculating a first optimal global histogram from said global histogram and said upper grey level threshold,
• means for calculating a second global histogram function from said lower grey level threshold,
• means for calculating a target sized histogram from said target sized histogram data and said low level threshold value, and
• look-up table means responsive to said upper threshold level global histogram said lower threshold level global histogram said target sized window histogram and said centroid for producing a probability of target function and a confidence function.