Method of identifying objects

• US 4,747,148 A
• Filed: 04/09/1984
• Issued: 05/24/1988
• Est. Priority Date: 04/11/1983
• Status: Expired due to Fees

• Current Assignee: Kabushiki Kaisha Komatsu Seisakusho
• Original Assignee: Kabushiki Kaisha Komatsu Seisakusho
• Inventors: Watanabe, Yuji, Kondo, Toshiaki
• Primary Examiner(s): Moore, David K.
• Assistant Examiner(s): Couso, J.
• Application Number:
  US06/598,379
• Time in Patent Office: 1,506 Days
• Field of Search: 901/46, 901/47, 382/8, 382/10, 382/18, 382/48, 382/1, 382/34, 382/42, 364/468, 358/101, 356/384, 356/385, 356/387, 356/394
• US Class Current: 382/141
• CPC Class Codes: G06V 10/20 Image preprocessing
• CPC Class Codes: G06V 10/32 Normalisation of the pattern dimensions

Abstract

A plurality of objects are identified by the steps of preparing a reference pattern B comprising picture image data of a predetermined number and representing the brightness relating to the configuration of the objects; extracting a brightness pattern A comprising a predetermined number of picture image data among picture image data in a predetermined field of view in which the objects presents; and determining the correlation between the reference pattern B and the brightness pattern A so as to judge as to whether the brightness pattern A is the brightness pattern related to the configuration of the objects. When determining the correlation, average values of respective patterns are subtracted respectively from the brightness pattern and the reference pattern to obtain patterns A'"'"' and B'"'"' obtained by converting then in connection with deviations from the average values. A pattern A" is obtained by normalizing pattern A'"'"' and the inner products of patterns A" and B'"'"' and of patterns A'"'"' and B'"'"' are determined. These two inner products are used as a reference value for identifying the objects.

Claims

1. A method of recognizing objects comprising the steps of:

extracting a brightness pattern A(Ca₁, Ca₂, . . . Ca_n) including n picture image data points from picture image data in a predetermined field of view in which a plurality of objects to be detected are present;

determining an average value Ca from said brightness pattern by using an equation ##EQU15## determining a pattern A'"'"' (C'"'"'a₁, C'"'"'a₂, . . . C'"'"'a_n) obtained by converting said brightness pattern in connection with a deviation from said average value according to an equation

space="preserve" listing-type="equation">C'"'"'ai=Cai-Ca;

determining a pattern A" (C"a₁, C"a₂, . . . C"a_n) obtained by normalizing said pattern A'"'"' according to an equation ##EQU16## preparing a reference pattern B (Cb₁, Cb₂, . . . Cb_n) comprising n picture image data points representing as brightness values the configuration of an object to be detected;

determining a pattern B'"'"' obtained by converting said reference pattern B in connection with a deviation from an average value of said reference pattern B;

calculating inner products of said patterns A" and B'"'"' and of said patterns A'"'"' and B'"'"', and

identifying said objects by using said products as evaluation values.

5. The method according to claim 1 wherein said identifying comprises determining whether an object is present by comparing said calculated inner products with preset threshold values.

2. In method of identifying objects which includes the steps of:

extracting a brightness pattern including a predetermined number n of picture element data which are continuous in a predetermined scanning direction from picture image data in a predetermined field of view in which a plurality of the objects to be identified are scattered;

repeating said extracting step for sequential sets of n picture elements each of which sets is offset from the preceding set by one picture element in said scanning direction; and

comparing each of said extracted brightness patterns with a prepared reference pattern representing the brightness of said predetermined number n of picture elements in a cross-sectional direction of said objects;

the improvement, for use where the brightness pattern in the cross-sectional direction of said objects varies in accordance with directions of said objects, comprising;

preparing brightness reference patterns in the cross-sectional direction of said objects which are present in said predetermined field of view, in each of four directions corresponding substantially perpendicularly to the horizontal, vertical, right 45°

upper and right 45°

lower directions;

extracting brightness patterns in each of four scanning directions of said horizontal, vertical, right 45°

upper and right 45°

lower directions in said field of view, from said picture image data in said predetermined field of view; and

utilizing one of said prepared four brightness reference patterns selected in accordance with the direction of scanning of said brightness pattern as a reference pattern for comparison with said brightness pattern, thereby facilitating the identification of randomly scattered objects.

3. A method of identifying objects comprising the steps of:

preparing a reference pattern representing brightness of a predetermined number of picture elements in a cross-sectional direction of said objects, said reference pattern being picture image data of said objects having a uniform cross-sectional configuration for a length of a predetermined section thereof;

extracting a brightness pattern of a predetermined number of picture elements which are present continuously in an X direction of rectangular coordinates from picture image data in a predetermined field of view in which said objects are present, and repeating said extracting step for sequential such brightness patterns, each such brightness pattern being shifted by one picture element in said X direction with respect to the prior extracted brightness pattern;

detecting an occurrence at which one of said objects is crossed in accordance with said brightness pattern and said reference pattern;

after detecting said occurrence, extracting three brightness patterns along a first scan line position displaced by one picture element in a positive Y direction of said rectangular coordinates with respect to the reference position at which the cross-section of said object is detected, said first brightness pattern being obtained at a position that is displaced by one picture elelment on a negative side of said X direction with respect to said reference position, a second brightness pattern being obtained at a position that is the same as said reference position, and a third brightness pattern being obtained at a position displaced by one picture element on a positive side with reference to said reference position;

selecting that brightness pattern most resembling said reference pattern from among said three brightness patterns;

determining, utilizing said selected brightness pattern, and storing the end position of said crossed object in said positive Y direction

processing the negative Y direction in the same manner as in said positive Y direction; and

storing the end position of said crossed object in said negative Y direction; and

determining a distance between said stored two end positions, and identifying that a proper object is present when said distance corresponds to the length of said predetermined section of said objects.

4. In method of identifying objects which includes the steps of:

preparing a reference pattern representing brightnesses of a predetermined number of picture elements in a cross-sectional direction of objects to be identified, said reference pattern being picture image data of an object having uniform cross-sectional configuration for a length of a predetermined section thereof;

extracting brightness patterns of a predetermined number of picture elements which are present continuously in an X direction of rectangular coordinates from picture image data in a predetermined field of view in which said objects are present, and repeating said extracting step for successive brightness patterns which are respectively shifted by one picture element in said X direction with respect to the prior extracted brightness pattern;

the improvement comprising;

detecting a position of crossing of one of said objects in accordance with said brightness patterns and said reference patterns, such detection requiring that a similarity evaluation value y_n between said brightness pattern and said reference pattern be above a certain threshold value y_n ;

tracing in a Y direction of said rectangular coordinates a position of crossing said objects by utilizing said detected crossing position as a reference position; and

while tracing, resetting said threshold value Yn as a function in which one or more previously obtained evaluation values y_n are the variables, and wherein

said reset threshold value Y_n is smaller than a preset constant threshold value Y_o when said one or more previously obtained evaluation values y_n are larger than said constant threshold value Y_o.

6. A method of recognizing objects comprising the steps of:

extracting brightness pattern including n picture elements from picture image data in a predetermined field of view in which one or more like objects to be detected are present;

determining an average value of said brightness pattern;

determining a pattern A'"'"' corresponding to the deviation of said brightness pattern from said average value;

determining a normalized pattern A" obtained by normalizing said pattern A'"'"';

preparing a reference pattern B comprising n picture element representing the brightness pattern of an object to be detected;

determining a pattern B'"'"' corresponding to the deviation of said reference pattern B from the average value of said reference pattern B;

obtaining a pattern similarity value by calculating the inner product of said normalized pattern A" and said pattern B'"'"';

obtaining a pattern strength value by calculating the inner product of said patterns A'"'"' and B'"'"', and

identifying said objects by using said pattern similarity value and said pattern strength value as evaluation values.

7. The method of claim 6 wherein said pattern B'"'"' is normalized to obtain a normalized reference pattern B" and wherein said pattern similarity value and pattern strength value are obtained respectively by calculating the inner product of said values A" and B" and the inner product of said values A'"'"' and B".