Back to results
US 2005/0238199 A1Published application
Object detection in electro-optic sensor images
AssigneeThe United States of America as represented by the Secretary of the Navy
InventorsAndrew Nevis, James Bryan, Brett Cordes, James Taylor, Mary Hulgan
Life of this patent
Estimating the term…
Abstract
A method of object detection in an image uses a background anomaly approach that searches for anomalies of a particular size and shape that are distinguishable from the image's local background. Included is a geometric classifier used to distinguish regularly-shaped objects from irregularly-shaped objects.
Claims15
A method of detecting objects in an image, comprising the steps of:
- providing an image defined by an array of pixels, each pixel from said array having an original value indicative of intensity;
- providing a first mask defined by two spaced-apart strip regions wherein a target region is defined between said two strip regions;
- applying, at each said pixel, a best fit criteria to a portion of said image defined by said first mask for a plurality of orientations of said first mask wherein said pixel is located within said target region for each of said plurality of orientations, said best fit criteria normalizing a mean square error of pixel intensities contained by said two strip regions by a mean square error of pixel intensities of said target region to generate a least squares error estimate for each of said plurality of orientations, wherein a best fit is indicated by a smallest least squares error estimate;
- generating, at each said pixel, an anomaly image defined by anomaly image pixels in correspondence with said array of pixels, wherein each said anomaly image pixel has a value defined as a difference between (i) said original value associated with a corresponding one of said array of pixels, and (ii) said smallest least squares error estimate;
- convolving said anomaly image with a second mask of a selected size and shape wherein convolution values are generated across said anomaly image, said convolution values that are greater than a threshold value being indicative of the presence of an object of interest in a region of said anomaly image, wherein one of said anomaly image pixels associated with a greatest one of said convolution values in said region is defined as a centroid of the object of interest;
- determining outer edges associated with the object of interest wherein a segmented object is defined by said outer edges with all pixels within said outer edges being assigned the same value;
- determining geometric features of said segmented object;
- and
- comparing said geometric features of said segmented object with selected threshold features to determine if said geometric features correlate with said selected threshold features.
6 more elements
A method of detecting objects in an image, comprising the steps of:
- providing an image defined by an array of pixels, each pixel from said array having an original value indicative of intensity;
- providing a first mask defined by two spaced-apart strip regions wherein a target region is defined between said two strip regions;
- applying, at each said pixel, a best fit criteria to a portion of said image defined by said first mask for a plurality of orientations of said first mask wherein said pixel is located within said target region for each of said plurality of orientations, said best fit criteria normalizing a mean square error of pixel intensities contained by said two strip regions by a mean square error of pixel intensities of said target region to generate a least squares error estimate for each of said plurality of orientations, wherein a best fit is indicated by a smallest least squares error estimate;
- generating, at each said pixel, an anomaly image defined by anomaly image pixels in correspondence with said array of pixels, wherein each said anomaly image pixel has a value defined as a difference between (i) said original value associated with a corresponding one of said array of pixels, and (ii) said smallest least squares error estimate;
- convolving said anomaly image with a second mask of a selected size and shape wherein convolution values are generated across said anomaly image, said convolution values that are greater than a threshold value being indicative of the presence of an object of interest in a region of said anomaly image, wherein one of said anomaly image pixels associated with a greatest one of said convolution values in said region is defined as a centroid of the object of interest;
- and
- determining size, shape and orientation of the object of interest using said centroid thereof.
4 more elements
Dependencies are inferred from the claim text.