# US 5,703,784 A · Machine vision apparatus and method for sorting objects

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 5,703,784 A
- Document ID: US5703784A
- Status: Granted
- Kind: A
- Application number: 08/550,310
- Filed: 1995-10-30
- Granted: 1997-12-30
- Earliest claimed: 1995-10-30
- Assignees: The United States of America as represented by the Secretary of Agriculture
- Inventors: Thomas C. Pearson
- Web page: https://patentgrep.com/patent/US5703784A

## CPC

- B07C 5/3422: {using video scanning devices, e.g. TV-cameras}
- B07C 5/366: {during free fall of the articles}
- G06V 10/44: Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- B07C: POSTAL SORTING; SORTING INDIVIDUAL ARTICLES, OR BULK MATERIAL FIT TO BE SORTED PIECE-MEAL, e.g. BY PICKING (specially adapted for a specific purpose covered by another class, see the relevant place, e.g. A43D33/06, B23Q7/12)
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

An object is sorted into different classes by producing a video image of the object. The signal level discontinuities in the image are evaluated by first deriving a video signal gradient of each pixel in the image. Then the number of pixels in three different signal gradient ranges are counted to produce count parameters which combined indicate the areas of differing contrast in the object. The count parameters are utilized in a pair of discriminate functions which define the two classes and solution of the discriminate functions are used to determine the probability that the object fits within one class. In response to the probability the object is directed along either a first path associated with one class or along a second path associated with the other class.

## Claims

### Claim 1

An object sorting method comprising steps of:
producing a video image of an object in which the video image comprises a series of pixels each having an intensity;
deriving signal gradients for a plurality of pixels in the video image;
defining a plurality of regions in the video image in which each region has a given plurality of pixels;
determining how many pixels have both a signal gradient that is less than a first signal gradient value and an intensity less than a first intensity value to produce a first parameter value;
determining how many regions exist wherein all the pixels in the region have both a signal gradient that is greater than a second signal gradient value and an intensity greater than a second intensity value to produce a second parameter value;
determining how many pixels have a signal gradient that is greater than a third signal gradient value and less than a fourth signal gradient value to produce a third parameter value; and
selecting a path along which to direct the object in response to the first, second, and third parameter values; and
directing the object along the selected path.

### Claim 2

A method of sorting an object into one of at least a first class and a second class, said method comprising steps of:
producing a video image of an object in which the video image comprises a series of pixels each having an intensity;
determining signal gradients for a plurality of pixels in the video image;
counting how many pixels have both a signal gradient that is less than a first signal gradient value and an intensity less than a first intensity value to produce a first parameter value;
defining a plurality of regions in the video image wherein each region has a given plurality of pixels; and counting how many regions exist in which all the pixels in the region have both a signal gradient that is greater than a second signal gradient value and an intensity greater than a second intensity value;
determining, from the first parameter count, a probability that the object should be classified within a first class;
selecting a path along which to direct the object in response to the probability; and
directing the object along the selected path.

### Claim 3 (depends on 2)

The method as recited in claim 2 wherein the step of counting further comprises counting how many pixels have a signal gradient that is greater than a first gradient value and less than a second gradient value.

### Claim 4

A method of sorting an object into one of at least a first class and a second class, said method comprising steps of:
producing a video image of an object in which the video image comprises a series of pixels each having an intensity;
determining signal gradients for a plurality of pixels in the video image;
counting how many pixels have both a signal gradient that is less than a first signal gradient value and an intensity less than a first intensity value to produce the first parameter count designated LFR; and further comprising:
defining a plurality of regions in the video image in which each region has a given plurality of pixels;
counting how many regions exist wherein all the pixels in the region have both a signal gradient that is greater than a second signal gradient value and an intensity greater than a second intensity value to produce a second parameter count designated HGR; and
counting how many pixels have a signal gradient that is greater than a third signal gradient value and less than a fourth signal gradient value to produce a third parameter count designated MG;
determining, from the first parameter count, a probability that the object should be classified within a first class;
selecting a path along which to direct the object in response to the probability; and
directing the object along the selected path.

### Claim 5 (depends on 4)

The method as recited in claim 4 wherein the step of determining a probability comprises:
utilizing parameter counts LFR, HGR and MG to solve a first discriminate function for the first class to produce a value D1;
utilizing the parameter counts LFR, HGR and MG to solve a second discriminate function for the second class to produce a value D2; and
calculating the probability as a function of the values D1 and D2.

### Claim 6 (depends on 5)

The method as recited in claim 5 wherein the first and second discriminate functions have the form:
D1 or D2=-k1+k2(MG)+k3(LFR)+k4(HGR)
where k1, k2, k3 and k4 are constants which have one set of values for the first discriminate function and another set of values for the second discriminate function.

### Claim 7 (depends on 5)

The method recited in claim 5 wherein the probability is derived from the expression: ##EQU2##

### Claim 8

A method of sorting an object into one of at least a first class and a second class, said method comprising steps of:
producing a plurality of video images of the object with each video image formed by a series of pixels each having an intensity;
determining signal gradients for a plurality of pixels in each video image;
for each video image counting how many pixels have both a signal gradient less than a first gradient value and an intensity less than a first intensity value to produce the first parameter count designated LFR; and further comprising:
defining a plurality of regions in the video image in which each region has a given plurality of pixels;
counting how many regions exists in which all the pixels in the region have both a signal gradient that is greater than a second gradient value and an intensity greater than a second intensity value to produce a second parameter count designated HGR; and
counting how many pixels have a signal gradient that is greater than a third gradient value and less than a fourth gradient value to produce a third parameter count designated MG;
determining, from the parameter count, a probability that the object should be classified within the first class; and
selecting a path along which to direct the object in response to the probability.

### Claim 9 (depends on 8)

The method as recited in claim 8 wherein the step of producing a plurality of video images comprising activating a plurality of video cameras that are placed at different angles with respect to a trajectory of the object.

### Claim 10 (depends on 8)

The method as recited in claim 8 wherein the step of counting further comprises adding the first parameter counts LFR for each image to produce a first sum LFRS; adding the second parameter counts LFR for each image to produce a second sum HGRS; and adding the third parameter counts MG for each image to produce a third sum MGS.

### Claim 11 (depends on 10)

The method as recited in claim 10 wherein the step of determining a probability comprises:
using sums LFRS, HGRS and MGS to solve a first discriminate function for the first class to produce a value D1;
using sums LFRS, HGRS and MGS to solve a second discriminate function for the second class to produce a value D2; and
deriving the probability as a function of values D1 and D2.

### Claim 12 (depends on 11)

The method as recited in claim 11 wherein the first and second discriminate functions have the form:
D1 or D2=-K1+K2(MGS)+K3(LFRS)+K4(HGRS)
where k1, k2, k3 and k4 are constants which have one set of values for the first discriminate function and another set of values for the second discriminate function.

### Claim 13 (depends on 12)

The method as recited in claim 12 wherein the probability is derived from the expression: ##EQU3##

### Claim 14

An apparatus for classifying objects, which apparatus comprises:
an input at which to receive a video image of an object in which the video image comprises a series of pixels each having an intensity;
a processor connected to said input and producing signal level gradients for a plurality of pixels in the video image;
an image evaluator coupled to said processor which comprises:
a first counter which determines how many pixels have both a signal gradient that is less than a first signal gradient value and an intensity less than a first intensity value to produce a first parameter value;
a second counter which determines how many regions exist wherein all the pixels in the region have both a signal gradient that is greater than a second signal gradient value and an intensity greater than a second intensity value to produce a second parameter value; and
a third counter which determines how many pixels have a signal gradient that is greater than a third signal gradient value and less than a fourth signal gradient value to produce a third parameter value; and
an output circuit which in response to the first, second and third parameter values transmits a command to a sorting device which command designates a path along which to direct the object.