US 4,244,650 AGrant
Automatic Optical Inspection and Sorting
Issue Date:1981-01-13
•4 Claims
•19 Drawing Sheets
Abstract
In this disclosure is described apparatus for automatic pattern or object recognition in which objects or goods with optically recognizable patterns are optically scanned for the pattern by passing the objects in front of optical sensor means where the light reflected from the object is received by the sensor means. In this disclosure there is described in detail apparatus for identifying and sorting various types of returnable bottles such as soft drink bottles by scanning across the bottles, the light reflected from the bottle graphics providing a coded signal or signature representing the bottle type. The identification is accomplished by considering such signature features as number of peaks, largest gap, area under curve, summation of difference between consecutive data points and the ratio of the area to the difference summation. The values obtained for the features are then matched with known feature bounds to provide the identification.
Metadata
Assignee
- Honeywell Inc.
Inventors
- James H. Garfunkel
- Kenneth P. Koeneman
Application Information
Application Number:US 898221&
Filing Date:1978-04-20
Priority Date:1978-04-20
Art Unit:257
Classifications
IPC:
G06K 938G06K 982G06K 946G01B 1100
Field of Search:
35625034071;446;448;394223 B;255;257;202146.3 Q;146.3 AC
Patent Drawings (19 sheets)
Description
Field of the Invention
This Invention relates generally to optical pattern recognition of goods such as bottles, labelled cans, fabric patterns and the like. In a specific embodiment described this invention relates to automatic identifying (sorting and separating) of different types of returnable bottles such as soft drink bottles. Refillable soft drink bottles which are returned to the bottler are previously sorted by size but not by type. This invention allows an automatic identifying by type.
Background of the Invention
With the increasing use of refillable containers, there is an expanding need in the soft drink industry for means of sorting bottles which are returned, so that the correct product goes into the proper bottle. At the present time, despite attempts to automate sorting in the past, sorting remains a manual operation at the bottling plant. Direct cost reduction to the bottler is possible by automating the sorting process. Examples of earlier patent art which have dealt with this problem in some form are U.S. Pat. Nos. 3,089,594, 3,228,520, 3,248,845, 3,257,897, 3,265,901, 3,351,198, 3,358,552, 3,529,169, 3,589,513, 3,955,179, 3,956,629 and 4,047,000.
Summary of the Invention
In the present invention which relates generally to optical pattern recognition of goods, an embodiment is described of a system which is capable of automatically identifying soft drink bottles in a non-contact sensing arrangement, without the need for a change in the bottle graphics. The sensing uses a scan line across the graphics on the bottle. The electro-optic sensing means comprises an light-emitting diode and silicon photo detector used in a reflective mode. The bottle is rotated in front of the sensor, and the different reflectance of the graphics (label) from that of the glass produces a varying signal which is characteristic of the type of bottle. A programmed microprocessor classifies the bottles as to type.
In the general operation of the bottle sorter apparatus, the individual bottles to be identified are moved in a single file by the bottle advancing conveyor into the inspecting area. At this point, each bottle is spun or rotated on its axis while the photo transistor senses the reflected radiation from the surface of the bottle for a full rotation and the bottle is identified. The identified bottle as it moves downstream from the inspecting station of the conveyor is then selectively shunted off the conveyor to the proper area for that type of bottle.
Brief Description of the Drawings
FIG. 1 is a much simplified block diagram of the system of the invention.
FIG. 2 is a diagrammatic representation of the sensor assembly.
FIG. 3 is a diagrammatic representation of a sample bottle to be identified with two sensors, one viewing reflections from the graphics of the bottle, and FIG. 4 is a graphical description of the analog output of the two sensors of FIG. 3.
FIG. 5 is a graphical description of the digitized signals of FIG. 4, and FIG. 5a is an expanded view of a section of FIG. 5.
FIG. 6 is a block diagram in more detail, and FIG. 7 is a schematic diagram of the electronic processing circuit.
FIG. 8 shows an example of identifying a bottle of the type shown in the previous figures.
FIGS. 9-12C are flow charts.
FIG. 13 shows the pattern received from several common bottles.
Description
In this invention, bottles are identified by direction light towards the bottle graphics, rotating the bottle 360.degree. and detecting reflected radiation, the reflected signature of the bottle providing identity. Particularly our identification of empty soft drink bottles is accomplished by analyzing the digital representations of discrete reflectance measurements in one or two circumferential bands at label level. The identification is accomplished by considering such signature features as number of peaks, largest gap, area under curve, summation of difference between consecutive data points, and the ratio of the area to the difference summation. The values obtained for the features are then matched with known feature bounds characteristic of each bottle to be identified.
Referring now to FIG. 1 which is a simplified functional block diagram, a bottle handling apparatus 10, which may comprise a conveyor system, transports a "BIG I" bottle 11 to a bottle identification area adjacent sensor means 12 and 13. BIG I is a simplified design for illustrative purposes. The sensor means each comprise a LED-phototransistor assembly. The LED (light emitting diode) illuminates the bottle adjacent the sensor with a beam of light. Light which is reflected from the bottle is sensed by the photo-transistor sensor which then provides an analog electrical output of a magnitude which is a function of the intensity of the reflected light. The bottle handling apparatus rotates the bottle while the bottle is in the identification area to produce an analog signature of the bottle reflection. The analog signal is connected to an analog-to-digital converter 14. The digital output signature is connected to recognition apparatus 15 which matches the features of the bottle with known feature bounds. The recognized bottle 11 is then directed to the proper location by sorting apparatus 16 which receives instruction from the recognition apparatus 15. Although in FIG. 1 the pattern to be recognized or identified is shown in the form of a bottle 11 which has an identifiable pattern thereon such as the bottle graphics, it is to be understood that the invention relates to optical pattern recognition generally and the object or goods having a pattern to be recognized is not intended to be limited to bottles and may include goods which are generally planar in surface as well as those which are like a bottle or a can, (i.e., generally or somewhat cylindrical). In FIG. 1 the item to be identified may be conveyed or translated past the sensor means without any rotation, if desired, as in the case of planar patterns. If a generally cylindrical item is to be identified it can preferably be rotated, or if it is properly oriented it can be conveyed past the sensor without rotation.
In one embodiment, each of the sensor means 12 and 13 comprises an assembly 18 shown schematically in FIG. 2 including an infrared LED 20, a focussing lens 21 for the LED, a photo-transistor detector 22 responsive to infrared, a focussing lens 23 for the photo-transistor, an infrared filter 24 to minimize the effect of ambient light or the photo-transistor. The assembly may be, for example, a Spectronics Type 1404-3 assembly. The LED 20 and the photo-transistor 22, together with their lenses are canted toward each other in the assembly so that the assembly has a common focal point of the LED and the photo-transistor about 0.2" in front of the assembly. The light source and the light detection may be separate elements, if desired, rather than an assembly.
Although the embodiment described for the sensor means is infrared responsive, it may be desirable when viewing reflection from certain colors of bottle graphics to utilize a different light such as a green LED a red LED or a yellow LED. The specified LED-phototransistor assembly described is illustrative and is not intended to be limiting. The use of the word light in this document is intended to include the infrared spectrum.
In FIG. 3 there is shown again the BIG I bottle 11 and the sensors 12 and 13 which provide analog outputs of the reflected signal when the bottle is rotated one revolution. The BIG I bottle 11 represents a bottle on which the only graphic feature (i.e. label) is a block I, which appears on opposite sides of the bottle and which graphic is seen only by the upper sensor. FIG. 4 shows graphically such a pair of signals with curve A representing the output of sensor 12 and curve B representing the output of sensor 13. FIG. 5 shows graphically the signals in digital form after being converted by the A-D converter, with curve A' being the digital form of curve A and with curve B' being the digital form of curve B. It will be noted that with the BIG I bottle the sensor 12 crosses the bottle graphics and sees the major reflected signal while sensor 13 sees essentially no signal.
FIG. 5 is shown with a vertical scale of 10 and a horizontal scale of 54. This is a somewhat simplified digital representation for descriptive purposes and is not intended to be limiting. In one successful embodiment of this invention the actual vertical scale is 0-31 and the horizontal scale is 512. In addition, the exact wave shape of the digital signal may vary from that shown in FIG. 5 and may be more like that shown in FIG. 5a which is an expanded and more exact representation of a portion of FIG. 5.
FIG. 6 discloses a more detailed functional block diagram of the overall system for sorting bottles. In this Figure, the bottle handling apparatus is shown as mechanical bottle manipulation apparatus 30 and a shaft encoder 31. The bottle manipulation apparatus includes bottle spinning means at the location of the sensors. In a laboratory prototype the bottle is placed in a fixture which causes rotation of the bottle by means of a D.C. motor, such as a Honeywell model HSM3D DC motor. The rotation rate is variable and is capable of spinning the bottle at rates greater than 1800 rpm. The two LED-phototransistor sensor pairs are mounted so that their height above the base and distance from the bottle may be adjusted. The sensor may be set to receive the reflections perpendicular to the bottle surface or may preferably be canted to view the approaching or receding bottle horizon. This canting of the sensors has the effect of eliminating much of the spiked pulses due to irregularities in the bottle glass occurring when the sensor is directly perpendicular to the bottle's surface. Both sensors in the preferred embodiment are canted in this fashion. In this way, the detector receives light which is diffusely reflected from the graphics, but not the light which is specularly reflected from the glass surface.
The shaft of the drive motor is linked to the bottle as well as optical shaft encoder 31. The particular encoder (a Hutchinson Industrial Corp. 2300 series shaft encoder) effectively divides the circumference of the bottle into 512 segments to provide 512 output cycles per revolution. Analog reflectance measurements are made when each pulse from the shaft encoder is output. These analog values are applied to an analog switch 32 which selects the particular sensor to be read. The shaft encoder pulse also starts the analog-to-digital conversion process at converter 14. When the process is finished, the A/D converter issues an "end of conversion" signal to a peripheral interface adaptor (PIA) 33.
A flag bit is set in the PIA 33 when "end of conversion" is indicated. The microprocessor 34 senses the condition of the flag, stores the digital representation of the reflectance value in random access memory (RAM) 35, and resets the flag. This process continues until all 512 values are stored in the RAM buffer.
When the buffer is full, the program uses the data in the buffer to isolate the features used for identification. When the features have been determined, the program proceeds to identify the bottle from the characteristics of the features. The other PIA 36 in the system issues the appropriate identification and commands to the bottle handling mechanism.
Obtaining and Storing Bottle Data
Bottle reflectance is measured by applying a fixed current to the infrared light emitting diode 20 which illuminates the bottle surface. The Darlington phototransistor sensor 32, which shares the same package 18 with the LED, provides an output voltage which is directly proportional to the bottle's surface reflectance. A filter 24 may be placed over the phototransistor aperture to remove most of the visible ambient light and its associated 60 Hz component where the 60 Hz or 120 Hz component is a source of unwanted signal.
In the present configuration, several sensors may be used and selected by the analog switch. In a current embodiment we are using data from two sensors 12 and 13. The analog switch 32 is now controlled by a flip-flop synchronized with the pulses from the shaft encoder 31 but may be controlled by the processor 34 when it is desirable to look at one particular sensor. We essentially interleave the data from each sensor in the memory buffer 35 alternately storing information from one sensor then the other for the entire circumference of the bottle.
This apparatus requires a time period of less than 10 usec for the A/D conversion, and with conversion rates of this order a successive approximation type of A/D converter is necessary. Specifically, in the present embodiment, we are using a 12-bit successive approximation A/D converter operating in the short cycle mode. Only the five most significant bits are presently used which gives a numerical range of values from 0 to 31 (0.0 volts to 5.0 volts in 0.15 volt steps). These five bits are stored in an eight bit data word in the five least significant bit positions. The most significant bit of the data word presently indicates the sensor source. The next two most significant bits are reserved for future use as sensor identifiers. When the bottle is rotating at 1800 rpm, each of the 512 segments is sampled for 65.12 usec yielding a 15.4 KHz sample rate for our system. The highest frequency in a waveform which can be reproduced qualitatively is the Nyquist frequency. This is equal to one half of the sampling rate, therefore the Nyquist frequency for our system would be 7.7 KHz. The Nyquist frequency as used in sampling theory is described, for example, in the text "Digital Processing of Signals" by Bernard Gold and Charles M. Rader, Lincoln Laboratories, MIT Series, McGraw Hill, 1969, pages 137-141. An optional low pass filter with a -3 dB frequency of no greater than 7.7 KHz may be used to remove high frequency components from the analog sensor data and still allow a faithful digital reproduction of the analog signal.
Description of Useful Bottle Signature Features
The bottle sorting system of this invention is based on analysis of relative signatures, in which relative signatures are defined as reflectance information read off the bottle starting at any random point. No absolute reference point is used to synchronize read out. The dimple (the molded indent) on the base of the bottle, for example, would be considered an absolute reference point. Using the relative signature allows bottle to be placed randomly under sensors and eliminates the need to orient the bottle before sensor data is read.
Primary bottle signature features to be described below in more detail are peak count, area under the signature, sigma value, largest gap, and ratio value (area/signal). Other features which also combine together with those above in determining the bottle signature include the peak spacing, the peak width (say at half-height), the number of points above some fixed number, the number of quartile crossings, the peak heights and of course, which of the detectors is giving the significant information.
Peak Count Feature
The bottle signature will appear to have a distinctive number of peaks due to the changes in reflectance as the bottle label passes under the phototransistor sensor. Each of the 512 reflectance values read from the bottle circumference will be converted to a digital form which will be any integer value from 0 to 31 depending on the reflectance. In the system design, we arbitrarily define a peak as any continuous or monotonic increase in ineger value for at least three consecutive readings followed by any continuous or monotonic decrease for at least three consecutive readings at any later point in time. Once these two conditions have occurred, a peak counter is incremented. The system continues looking for the next three increases followed by three decreases until the entire buffer has been examined, and the counter provides the number of peaks in the bottle signature.
Referring back to FIG. 5 which represents the digital signal of the BIG I bottle, it can be seen that there are two peaks in the signature as measured by sensor 12. The monotonic increases at a, b, c, and d are followed by monotonic decreases at e, f, and g thereby meeting the definition for a peak. The second peak is similar in nature.
Area Under Signature Feature
The area feature is simply the sum of the integer values for all data points associated with a particular sensor (i.e., the total area under the curve). In the example of FIG. 5, the area for sensor 12 is 68 and the area for sensor 13 is 6.
Sigma Feature
The sigma feature is the sum of the absolute values of the differences between consecutive integer data values for all the points associated with a particular sensor (i.e. absolute value of the sum of the vertical excursions). In the example of FIG. 5, the sigma feature is 40 for sensor 12 and 4 for sensor 13.
Largest Gap Feature
A gap is defined as the number of consecutive data points which are less than or equal to the integer 2 in the range of 0-31 (i.e., the length of the largest gap between peaks). The threshold value, 2 is an arbitrary design choice and may be changed if necessary.
Ratio Feature
The ratio feature value is the area feature value divided by the sigma feature value.
FIG. 7 shows a schematic representation of an electronic interface for digitizing the signals received from the sensors. The signal from the shaft encoder 31 which is brought in on line 40, drives a flip-flop 41, such as a 7474, which in turn actuates an electronic switch 42, here shown as an RCA type CD4016. The analog signal from sensor 12 and the signal from sensor 13 are amplified by conventional 741 op. amps. 43 and 44, respectively, and connected by conductors 45 and 46 to the inputs of the switch 42. Switch 42 alternately passes the signal from sensor 12 and sensor 13, so that for each bottle rotation, the output consists of 256 pulses from sensor 13 interleaved with 256 pulses from sensor 12. The signals from switch 42 are amplified 47 and then are connected by conductor 50 into the analog-to-digital converter 14, such as Burr Brown type ADC80AG. The encoder signal on line 40 is also shaped by signal shapers 51 and 52 (type 74123) connected to the converter 14 and used as the convert command signal input to the A/D converter. The digitized output of converter 14 is brought out through a plug which provides proper input connections for the microcomputer.
A simplified explanation of the way the signals are utilized in the microcomputer is shown diagramatically in FIG. 8. First the outputs from the two sensors are checked for the area; the one with the largest area (in this case #1) is picked and further analyzed. The sensor which is scanning a label or other identifying marks will generally have a much larger area under the signature with respect to the sensor which is just scanning bottle surface (no label) only. Therefore, the system's first consideration will be to select the sensor which is scanning the surface with valid identifying markings.
The next step consists of comparing the gap, peak count and ratio value with predetermined ranges and assigning a unique bit pattern to the range in which the value falls. This bit pattern is stored in the appropriate area of the ID.word. The ID word is masked and modified to remove "don't care" states then compared with a known pattern to achieve bottle identification. Said another way, the number of peaks 54, longest gap 55 and the ratio of area to .SIGMA. 56 are computed, the ratio area/.SIGMA. is used in order to arrive at a "normalized" number, since both features are affected by the brightness of the graphics. These three features are then catagorized into groups, and the bottle identified by the combination. In the case of the BIG I bottle, the two peaks provide an output from the peak counter 54 on the 0.ltoreq.p.ltoreq.3 output. The means 55 for determining the longest gap provides an output on the 17.ltoreq.G.ltoreq.30 output; and the means 56 determining area divided by sigma provides an output at 0.ltoreq.R.ltoreq. 4. The BIG I bottle can thus be identified.
Referring to FIGS. 9-12C, the flow diagram form of the computations performed in identifying the bottles is described. In FIG. 9, which shows a generalized flow, the program is entered through a system turn-on sequence and passes to the START-UP subroutine which initializes the system, the initialization procedure being known to those of ordinary skill in the microprocessor art and not being of significance in the invention. Following the START-UP control passes A to the BOTTLE CONTROL step where the determination is made has bottle control status indicated a ready condition. If not, the next step investigated is to determine whether the time to ready condition has become excessive, and if not the program returns to the BOTTLE CONTROL to test again. When the status indicates a ready condition, the program proceeds to call subroutine AA which reads in data from the sensors and stores the data in the buffer storage. Next the subroutine BB is called which determines the gap width, peak count, sigma, and the ratio of area/sigma using the data which was previously stored in the buffer. Then the subroutine CC is called which identifies the bottle, this by using the features found in the immediate preceding subroutine. The program then proceeds to inquire whether the bottle has been identified and if so proceeds to issue identification for display or bottle control and sends "ready for next bottle" signal.
FIG. 10 is a flow diagram of the subroutine AA which reads in data from the sensors. When a GO signal is received the program proceeds to initialize the input/output ports and then proceeds to test whether the sensor output has been converted to a digital word. If the test is yes the program proceeds to reset the word ready flag and to store the data word in the buffer and then to compare the current buffer address with the end address. The program then tests whether the buffer is full, if not the buffer address counter is incremented and the program returns to test for the next digital word. When the buffer is full, the subroutine ends and the subroutine BB is called up.
FIGS. 11a-11g show the flow diagram of the subroutine BB of FIG. 9, which manipulates the data in the buffer to obtain the "area" under the signature, the largest "gap" in the signature, "sigma" (summation of the absolute values of the differences between successive data values of signature), and the "number of peaks" in the signature. In FIG. 11a the subroutine is entered, the first step is to reset the counter locations. The stack pointer is saved thus preserving the address of the buffer stack. The first pass flag is then set and the program proceeds to put the starting address of the buffer in the stack pointer. The procedure advances on the flow diagram through point B and proceeds to get a data word. In the present description this may be the first word of the 512 stored in the buffer (or may be succeeding words triggered from the indicated FIG. 11d input of point B). In the process of retrieving the word, the stack pointer is also incremented to the next data word. The program then proceeds to test which sensor the data word is from. Continuing on FIG. 11b, the program proceeds to set up the area parameter by loading words into temporary locations which pass the values to be manipulated to the area subroutine. At this point in FIG. 11b the program jumps to an indicated subroutine which determines the signature area. First the data goes into the A-Register; the data word is saved and the program proceeds to clear the carry flag. Next the B-Register is loaded with the least significant 8 bits of the area word. The subroutine then adds the A and B registers and saves the result in the area word. Next the subroutine tests if the carry flag has been set, and if so, proceeds to increment the most significant 8-bit word in the 16-bit pair then saves the 16-bit area word. If the carry flag has not been set the original data is restored in the A-register. Now program flow returns out to the main program and saves the area word (16 bit).
In FIG. 11c the program proceeds to set up the GAP parameters and jumps to the GAP routine. In the GAP subroutine the first test determines if the data is greater than a predetermined threshold. If not, the lower-than-threshold counter is incremented to provide the most recent gap width count. The next step is to determine if the present threshold counter is greater than that previously stored gap width. If not, return from subroutine, if yes, increment the stored gap width and return from subroutine. If the first test determines the data is greater than the threshold, the subroutine proceeds to determine if this is the first time the data has become greater than the threshold. If yes, the initial gap is stored for later addition to the final gap. The initial gap-final gap situation is also known as a split gap condition and occurs because the bottles are not indexed, but begin the rotation at a random point on the circumference of the bottle. Then the subroutine clears the first gap indicator word, to clear the most recent gap width count and returns from subroutine. If this is not the first time the data has become greater than the threshold, the subroutine clears the most recent gap width count and returns from subroutine. The program then proceeds to update the gap parameters.
In FIG. 11d, the program first tests whether it is the first pass through the main program and if so, sets the first pass flag, proceeds to move the data to the reference position, passes junction Y and tests if entire buffer has been read. If entire buffer has not been read, go back to B, FIG. 11a. If on the second or successive pass, the next test determines if the data word equals the reference word. If it does not, there is a vertical excursion in the signal level which will provide an addition to the sigma total. Thus, if the data word does not equal the reference word, the next step is to save the data and the reference words, then to set up sigma parameters and jump to the sigma subroutine. Since the sigma determining subroutine is shown in FIG. 11e, a brief reference to that figure will be made before returning to the subroutine of FIG. 11d. In FIG. 11e, the first step is to clear the carry bit (in condition code register) followed by a test which determines if the data is equal to the reference. If not, a test is made to determine whether the data is greater than the reference. If it is, the next step is to subtract the reference from the data. If not, the data and reference are interchanged to make the result of the subtraction positive and then the subtraction is executed. In either case, the resulting difference from the subtraction step is added to the least significant 8 bits of sigma. Then the subroutine checks if the carry bit has been set. If no, return from subroutine to FIG. 11d. If yes, increment the most significant 8 bits of sigma, clear the carry bit and return from subroutine to FIG. 11d.
Back in the flow chart of FIG. 11d, the next step is to save the updated sigma values and to restore the data and reference if they were reversed above. The program then proceeds to determine if the data has increased X number of times. (3 was arbitrarily selected as X). If not, test if the data is greater than the reference. If so, increment the "data increasing counter" and check again if data has increased X consecutive times. If yes, set the "data increasing counter" flag and go to Y. If the data was not larger than the reference, then the "data increasing counter" is reset to zero and go to Y. If the program had determined earlier that the data had increased X number of times, then the next step is to determine if data is less than reference. If no, reset the "consecutive decreasing counter" and go to Y. If yes, increment the "consecutive decreasing counter" and determine if data has decreased X consecutive times, if not, go to Y. If yes, increment the peak counter to indicate another peak, reset the increasing and the decreasing consecutive counters and go to Y. When the entire buffer has been read, the output at the bottom of FIG. 11d is at G and the flow chart continues at FIG. 11f. FIG. 11f shows the subroutine which adds the initial gap to the final gap if such a split gap situation exists. The first step is to clear the overflow bit flag. Then an addition step adds the count in the "lower-than-threshold-counter" to the split value obtained earlier and discussed in FIG. 11c. Then the subroutine proceeds to determine if the overflow bit has been set. If yes, the next step is to replace the gap width count with a count of 255 which indicates the gap is the entire bottle circumference, and then return from subroutine. If not, the next step determines if the previous addition sum is larger than the current gap width count. If not, return from subroutine; if yes, replace the gap count with the new value resulting from the addition and then return from subroutine.
In FIG. 11g, the program continues entering at H and jumps to the subroutine which calculates the ratio between area and sigma. The area is loaded into the A and B registers (dividend), sigma is loaded into the X-register (divisor) and the division is performed. The ratio is stored in the A and B registers, the original stack pointer is restored and program flow advances to the program which identifies the bottle.
The flow diagram of FIGS. 12a, 12b and 12c describe in more detail the subroutine CC of FIG. 9 which identifies the bottle using features found in the previous subroutine. In FIG. 12a after initializing, the first test is to determine which sensor indicated the largest area under the curve, then the sensor number and the largest area is stored. The program then proceeds to set bit 14,15 of ID word with the appropriate sensor number and to set bit 12 of the ID word if the area is greater than a predetermined number. The program then proceeds to transfer the ratio, the peak count, and the gap to temporary working locations. Continuing on FIG. 12b, the program then proceeds to assign bits 8,9, 10 and 11 a pattern to correspond with a particular band of ratios, to assign bits 4, 5, 6 and 7 a pattern to correspond with a particular band of peaks, and to assign bits 0, 1, 2 and 3 a pattern to correspond with a particular band of gap widths. The program then proceeds to save the unmodified coded ID word and to clear a word location into which the decoded ID word will be stored. In FIG. 12c, the coded word is then masked to remove "don't care" situations. The ID word is then compared with standard codes which have been derived for different types of bottles and when a comparison occurs to issue the decoded ID word. If a comparision does not occur, the program sets the decoded ID word to an all-zero condition. The program then exits to bottle handling.
In our design and testing up to the present time we have been able to identity 11 types of soft drink bottles using two sensors in the apparatus described. They are: COKE, SPRITE, FRESCA, TAB, 7-UP. DR. PEPPER, PEPSI, CRUSH, MASON'S, MOUNTAIN DEW, TOM MOORE, and in addition among these bottle types, three of them have two variations of graphics and two others have three variations of graphics. Sample signatures obtained from several types of bottles as they appear on an oscilloscope are shown in FIG. 13.
A program which is written in basic languages for directing the microprossessor operation to identify and sort bottles forms a portion of the description as set out in the following table. ##SPC1## ##SPC2## ##SPC3##
Claims
The embodiments of the invention in which an exclusive property or right is claimed are defined as follows:
1. Pattern recognition apparatus comprising: optical sensor means detecting reflected light from the surface of goods having a pattern to be identified, the sensor means scanning a line across the goods as the goods pass in front of the sensor means, the light reflected from the pattern of the passing goods providing an analog signal pattern output from said sensing means which analog signal pattern is characteristic of the pattern sensed; means for sampling the analog signal pattern at predetermined increments of movement of the goods and means converting each said signal sample to a digital bit; means for temporarily storing the digital bits representing the signal pattern from said goods, said temporarily stored digital signal containing information which can be expressed as signal peaks, area under the curve of the digital signal, gap in the signal, and vertical excursions of the digital signal from one bit to the next; means for counting the number of peaks of the digital signal; means for determining the area under the curve of the digital signal; means for determining the largest gap in the digital signal, where gap is defined as the number of consecutive data points at a signal level of less than a predetermined digital number; means for summing the total of the positive and negative vertical excursions of the digital signal, the sum being referred to as sigma; means for dividing the area value by the sigma value to obtain a value Y; and, means for comparing the number of peaks, the area, the largest gap, and the value Y with prestored information characteristic of each pattern type to be recognized.
2. Bottle identifying apparatus comprising: means for rotating a bottle to be identified; optical sensor means detecting reflected light from the surface of a bottle to be identified, the sensor means scanning a line across the graphics of the bottle as the bottle rotates in front of the sensor means, the light reflected from the rotating bottle graphics during a rotation of the bottle providing an analog signal pattern output from said sensing means which analog signal pattern is characteristic of that bottle type; means for sampling the analog signal pattern at angular increments of 360.degree./x during the rotation of the bottle and means converting each signal sample to a digital bit where x is an integer; means for temporarily storing the x digital bits representing the signal pattern from a rotation of said bottle, said temporarily stored digital signal containing information which can be expressed as signal peaks, area under the curve of the data points of the signal, gap in the signal, and vertical excursions of the digital signal from one bit to the next; means for counting the number of peaks of the digital signal; means for determining the area under the curve of the digital signal; means for determining the largest gap in the digital signal, where gap is defined as the number of consecutive data points at a signal level of less than a predetermined digital number; means for summing the total of the positive and negative vertical excursions of the digital signal, the sum being referred to as sigma; means for dividing the area value by the sigma value to obtain a value Y; and, means for comparing the number of peaks, the area, the largest gap, and the value Y with prestored information characteristic of each bottle type to be recognized.
3. Bottle identifying apparatus, comprising: first and second optical sensor means; means for rotating a bottle to be identified, said first and second optical sensor means each being mounted in proximity to the rotating bottle and each illuminating an area of and receiving reflective light from bottle graphics in the illuminated area of the rotating bottle, the sensor means each providing an analog electrical output signal proportional to the reflected light from the bottle surface as the bottle rotates; shaft encoder means connected to said bottle rotating means, said shaft encoder means providing a predetermined number of output electrical pulses during a rotation of the bottle; means including switching means actuated in response to the shaft encoder pulses and selectively connecting said first and second sensor means output sigma to analog-to-digital converter means at each occurrence of said shaft encoder means; analog-to-digital converter means actuated in response to the shaft encoder pulses having an input circuit adapted to be connected to receive said analog signals and to convert said signals to a digital form; means connecting said shaft encoder output pulses to said converter means to initiate the analog-to-digital conversion at the occurrence of each said pulse; permanent storage means for storing digital information characteristic of each bottle type to be recognized; memory means for temporarily storing the digital bits representing the two signal patterns from a rotation of said bottle, said temporarily stored digital signals containing information which can be expressed as signal peaks, area under the curve of the digital signal, gap in the signal, and vertical excursions of the digital signal from one bit to the next; means for determining the area under the curve of the digital signal; means for counting the number of peaks of the digital signal; means for determining the largest gap in the digital signal, where gap is defined as the number of consecutive data points at a signal level of less than a predetermined digital number; means for summing the total of the positive and negative vertical excursions of the digital signal, the sum being referred to as sigma; means for dividing the area value by the signal value to obtain a value Y; and, means for comparing said number of peaks, said area under the curve, said largest gap, and the value Y with prestored information as to peaks, area, gap and Y characteristic of each bottle type to be recognized.
4. The apparatus of claim 3 wherein said means for comparing further comprises means for checking the digital signal outputs of the two sensors for the area under the curve of each and for selecting the signal of larger area for further analysis.