Method of identifying an object in a visual scene

Abstract

A profile of an object is extracted from an image of a visual scene and a center of the profile is then determined. A measurement vector is determined comprising a plurality of distances to the profile from the center, at a plurality of angles relative to and through the center. The measurement vector is filtered with a central moving average filter, and then compared with at least one stored vector representative of at least one type of reference object, wherein the central moving average filter is responsive to the type of the at least one type of reference object, responsive to which a determination is made as to whether or not the object corresponds to at least one reference object, so as to provide for identifying the object in the visual scene.

What is claimed is: 
     
         1 . A method of identifying an object in a visual scene, comprising:
 a. extracting at least a profile of an object from an image;   b. determining a center of said profile;   c. determining a measurement vector comprising a plurality of distances from said center to said profile at a plurality of angles relative to and through said center;   d. filtering said measurement vector with a central moving average filter;   e. comparing said measurement vector with at least one stored vector representative of at least one type of reference object, wherein said central moving average filter is responsive to said type of said at least one type of reference object; and   f. determining whether or not said object corresponds to at least one said reference object responsive to the operation of comparing said measurement vector with said at least one stored vector.   
     
     
         2 . A method of identifying an object in a visual scene as recited in  claim 1 , wherein the operation of determining said center of said profile comprises determining a center-of-mass of said profile. 
     
     
         3 . A method of identifying an object in a visual scene as recited in  claim 1 , wherein each distance of said plurality of distances is determined by first radially searching radially outwards along a search direction from said center until reaching a point on said profile, and then calculating said distance from said center to said point on said profile. 
     
     
         4 . A method of identifying an object in a visual scene as recited in  claim 1 , further comprising deriving said at least one stored vector from a plurality of associated measurement vectors from a corresponding plurality of different images of a corresponding said reference object. 
     
     
         5 . A method of identifying an object in a visual scene as recited in  claim 1 , wherein said at least one stored vector is responsive to different images of a corresponding said reference object are at corresponding different ranges to said corresponding said reference object. 
     
     
         6 . A method of identifying an object in a visual scene as recited in  claim 4 , wherein said at least one stored vector is derived as a median value of said plurality of associated measurement vectors from said corresponding plurality of different images of said corresponding said reference object. 
     
     
         7 . A method of identifying an object in a visual scene as recited in  claim 4 , wherein the operation of deriving said at least one stored vector further comprises filtering the data of said plurality of associated measurement vectors with said central moving average filter prior to storing said at least one stored vector. 
     
     
         8 . A method of identifying an object in a visual scene as recited in  claim 7 , wherein a length of said central moving average filter is dependent upon said type of said corresponding said reference object. 
     
     
         9 . A method of identifying an object in a visual scene as recited in  claim 8 , wherein said length of said central moving average filter is the same for both said measurement vector and said at least one stored vector. 
     
     
         10 . A method of identifying an object in a visual scene as recited in  claim 9 , wherein said at least one stored vector comprises a plurality of stored vectors and said at least one type of reference object comprises a plurality of types of reference objects of a variety of object classes. 
     
     
         11 . A method of identifying an object in a visual scene as recited in  claim 10 , wherein said length of said central moving average filter is configured so as to substantially maximize a count of intra-class stored vectors of said plurality of stored vectors corresponding to said corresponding said reference object at different ranges, that are relatively strongly cross-correlated with one another. 
     
     
         12 . A method of identifying an object in a visual scene as recited in  claim 1 , wherein said at least one stored vector comprises at least two stored vectors, wherein a first of said at least two stored vectors is derived from at least one image of said reference object, and a second of said at least two stored vectors is derived from at least one mirror image of said reference object, wherein each said at least one mirror image is a left-right mirror image of a corresponding said at least one image. 
     
     
         13 . A method of identifying an object in a visual scene as recited in  claim 1 , wherein the operation of comparing said measurement vector with said at least one stored vector is responsive to corresponding portions of said measurement vector and said at least one stored vector that span less than 360 degrees between associated respective tails of said measurement vector and said at least one stored vector, respectively. 
     
     
         14 . A method of identifying an object in a visual scene as recited in  claim 13 , wherein each said at least one stored vector and said measurement vector are determined at said plurality of angles that are substantially equally distributed over a span of 360 degrees, and said at least one stored vector and said measurement vector are each subsequently filtered with said central moving average filter prior to the operation of comparing said measurement vector with said at least one stored vector. 
     
     
         15 . A method of identifying an object in a visual scene as recited in  claim 1 , wherein the operation of comparing said measurement vector with said at least one stored vector comprises cross-correlating said measurement vector and said at least one stored vector so as to generate a resulting cross-correlation value, and the operation of determining whether or not said object corresponds to said at least one said reference object is responsive to a comparison of said resulting cross-correlation value with a threshold.