Fuzzy logic technique to determine search time and probability of detection for targets of interest in background scenes

• US 6,424,736 B1
• Filed: 05/24/1999
• Issued: 07/23/2002
• Est. Priority Date: 05/24/1999
• Status: Expired due to Fees

• Current Assignee: Army Secretary of The United States of America Government As Represented By The
• Original Assignee: The United States Of America As Represented By The Secretary Of The Army
• Inventors: Singh, Harpreet, Meitzler, Thomas J., Sohn, Euijung
• Primary Examiner(s): Boudreau, Leo
• Assistant Examiner(s): CHAWAN, SHEELA C
• Application Number:
  US09/332,251
• Time in Patent Office: 1,156 Days
• Field of Search: 382/103, 382/106, 382/155, 382/159, 382/181, 382/224, 382/228, 706/1, 706/2, 706/9, 706/45, 706/900, 706/905, 701/27
• US Class Current: 382/155
• CPC Class Codes: G06F 18/21 Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
• CPC Class Codes: G06V 10/255 Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
• CPC Class Codes: Y10S 706/905 Vehicle or aerospace

Abstract

A method of determining the visibility of a target in a background uses search time as the output. A set of images of the target in the background is made, and selected input variables in each image are measured. The input variables relate to target size, target juxtaposition relative to the viewer'"'"'s location, luminance of the target and luminance of the background scene. One version of our method additionally uses wavelet edge points as an input variable. Each input variable, as well as the output variable, has several applicable membership functions by which membership values are assigned to the variables. Choosing membership functions for variables is done by fuzzy rules. Applying the fuzzy rules and membership functions produces multiple values for the output variable. These values are defuzzified to obtain a crisp end result. This result can disqualify proposed target designs or to help select among several good designs of the target.

Claims

1. A method incorporating fuzzy logic techniques to determine the degree of difficulty in finding a target of interest within a scene, wherein the difficulty is measured in terms of an output variable defined as search time, the method comprising:

making a set of images of the scene containing the target;

determining the value of a set of input variables of the scene wherein the input variables include selected variables relating to target size, target juxtaposition relative to a point of view, luminance of the target and luminance of the scene;

creating sets of first membership functions, one of the sets being for each of the input variables, the first membership functions associating the input variables with membership values;

creating a set of second membership functions, the second set being associated with the output variable;

creating a set of fuzzy rules, wherein for every fuzzy rule a particular membership function from each of the sets of membership functions is a selected function, whereby each fuzzy rule is comprised of a group of the selected functions;

after the values of the input variables are determined, then for each fuzzy rule determining the membership values for the selected functions from the first sets, and then determining average membership values for the selected membership functions from the first sets;

using the average membership values and associated selected membership functions from the second set, determine an output value for each fuzzy rule; and

using the output values to calculate a crisp value.

2. A method incorporating fuzzy logic techniques to determine the degree of difficulty in finding a target of interest within a scene, comprising:

making a set of images of the scene containing the target;

determining the value of a set of input variables of the scene wherein the input variables include selected variables relating to luminance of the overall scene, target, and vicinity of the target, as well as the juxtaposition, height, area, and distance of the target relative to a point of view within the scene;

creating sets of first membership functions, one of the sets being for each of the input variables, the first membership functions associating the input variables with membership values;

creating a set of second membership functions, the set of second membership functions being associated with the output variable;

creating a set of fuzzy rules, wherein for every fuzzy rule a particular membership function from each of the sets of membership functions is a selected function, whereby each fuzzy rule is comprised of a group of the selected functions;

after the values of the input variables are determined, then for each fuzzy rule determining the membership values for the selected functions from the first sets, and then determining average membership values for the selected membership functions from the first sets;

using the average membership values and associated selected membership functions from the second set, determine an output value for each fuzzy rule;

using the output values to calculate a crisp value; and

based on the crisp value, accepting or rejecting the target as a design candidate.

3. A method incorporating fuzzy logic techniques to determine the degree of difficulty in finding a target of interest within a scene, comprising:

making a set of images of the scene containing the target;

determining the value of a set of input variables of the scene wherein the input variables include selected variables relating to target size, target juxtaposition relative to a point of view, luminance of the target and luminance of the scene, and wherein the selected variables specifically include target distance, target aspect, target height, target area, target luminance, overall scene luminance, and luminance of target vicinity;

creating sets of first membership functions, one of the sets being for each of the input variables, the first membership functions associating the input variables with membership values;

wherein the first membership functions correspond to input function graphs on a first two-dimensional coordinate system, one dimension of the first system being membership value and ranging from 0 to 1, and another dimension of the first system being a range of values for one of the input variables, the input function graphs all having trapezoidally shaped portions in parts of the ranges where their membership value is greater than 0, and flat apexes of the trapezoidally shaped portions having the membership value of 1;

creating a set of second membership functions, the second set being associated with the output variable;

wherein the second membership functions correspond to output function graphs on a second two-dimensional coordinate system, one dimension of the second system being membership value and ranging from 0 to 1, and another dimension of the second system being a range of values for the output variable, the output function graphs all having trapezoidally shaped sections in regions of the ranges for the output variable where their membership value is greater than 0, and flat apexes of the trapezoidally shaped portions having the membership value of 1;

creating a set of fuzzy rules, wherein for every fuzzy rule a particular membership function from each of the sets of membership functions is a selected function, whereby each fuzzy rule is comprised of a group of the selected functions;

after the values of the input variables are determined, then for each fuzzy rule determining the membership values for the selected functions from the first sets, and then determining average membership values for the selected membership functions from the first sets;

using the average membership values and associated selected membership functions from the second set, determine an output value for each fuzzy rule; and

using the output values to calculate a crisp value.

4. The method of claim 3 wherein:

the input function graphs have intersections with one another, the intersections lying at one end of non-zero segments of the input function graphs;

all points on the non-zero segments have the same membership value; and

the input function graphs have other segments between the intersections and the apexes, the other segments and the apexes all having a membership value of 1.

5. The method of claim 3 wherein the set of input variables includes the number wavelet edge points in a scene determined from one or more derivatives of a wavelet transform of a function associating pixel location with gray scale value in the images.