JP3358142B2 - Object identification system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an object identifying apparatus capable of recognizing an object moving at a high speed and selecting the object at a high speed.

[0002]

2. Description of the Related Art Conventionally, for example, an image processing technique is used to confirm and inspect that soy sauce and mustard attachments are always inserted into a natto pack.

In other words, such an inspection apparatus detects a pack of natto flowing without irregularly arranging on a belt conveyor with a black-and-white camera, and attempts to recognize the pack using a grayscale image.

In such measurement processing, the distinction between natto, soy sauce, and mustard vinyl packages is made by image processing technology of density conversion. This pack of natto can be up to 3 per second
Since 0 pieces flow on the belt conveyor without being arranged irregularly, such a natto accessory identification device needs to detect and identify the accessory of the natto pack at high speed,
Monochrome image processing with a short processing time has been tried.

[0005] In addition, the use of color image processing technology for confirming and inspecting beer bottle caps has been tried, but in this case, only two or three bottles can be recognized per second, which is not practical. Currently, recognition is performed using black and white images.

[0006]

However, since the density of natto and mustard is similar in the apparatus for recognizing grayscale images as described above, the shape and existence of mustard package cannot be recognized, and it is practical. Absent. In addition, the detection rate of the cap of the beer bottle is not high, and manual work has occurred.

In order to increase the recognition rate, it is considered effective to perform not only shape feature extraction using a density conversion processing technique using a monochrome image but also color feature extraction. However, a conventional image processing apparatus that can be installed in a factory requires a great deal of time to process color information. Therefore, conventionally, in order to perform pattern measurement processing at high speed, when performing graphic processing of an image, a plurality of pixel processors (processors that handle pixels (pixels) constituting the screen) transfer the screen in the horizontal and vertical directions. A method of allocating a small area, which can be divided into two, as a unit, a method of allocating in units of scan lines, and a method of allocating in units of pixels are employed.

According to these allocation methods, the number of divisions is increased to reduce the area allocated to the pixel processor, thereby reducing the memory capacity of each processor.
Further, the processing time by the pixel processor is reduced in proportion to the number of processors.

However, in such a graphic processing method, since the processing of one screen is simply divided and processed, the overall memory capacity does not change, and the memory capacity and the time required for unnecessary access are not improved. However, there is a problem that the circuit configuration becomes complicated as the number of devices increases.

[0010]

Means for Solving the Problems] Therefore, in the present invention, transportable
The object conveyed by the feeding means is captured by the solid-state image sensor camera
An object detection unit for detecting and outputting image data;
Input for digitally converting the image data output from the output unit.
Input unit and arithmetic processing of image data input from the input unit
Image processing unit for performing the image processing, and the image data processed by the image processing unit.
The system control unit that performs processing of an object body identified for over data
And the operation of the input unit, the image processing unit, and the system control unit.
And a host computer that controls each operation.
This is an object identification device.

In particular, the input unit outputs from the object detection unit.
A detection unit control unit for capturing the image data, and a first data
Control from digital signal processor unit
Output the address information of the image data based on the
Memory address controller and the same input
・ Address output from memory address controller
Based on the address information,
Digitizes the data of the address information corresponding to the image data.
With an A / D converter that converts the
is there.

[0012] The image processing unit is inputted from the input unit.
Block memory to temporarily write the digital amount of image data
Memory and image data written to the same block memory
Digital signal processor that performs arithmetic processing based on
Sessa and the operation of the digital signal processor
Plane memory used in
Digital signal processor based on data control
And a controller for controlling the second digital
.At least two signal processor units
Digital signal processor configured as one set
It has a plurality of monitor unit sets.

Further, the system control unit includes an image processing unit.
Each digital signal processor unit set and
Third digital signal processor unit connected
And the third digital signal processor unit
Connected to the first digit array network and the first digital
Object that controls the signal processor unit
Includes a digital signal processor unit for identification
It is provided.

Further , the input unit, the image processing unit, and the system
Each digital Shea Gunaru processor provided in the arm control unit
Each unit performs local parallel pipeline processing
Processor, which are controlled by the host computer.
Are operated in parallel.

In the input section, the output from the object detection section
Analog image data of red and green
Data into the detector control unit as blue data,
By the first digital signal processor unit
Input memory address control based on control
Based on coarse pixel address information output from roller
Red data, green data, blue
A / D converter converts color data into digital quantities
Converts and outputs the image.
Red, green and blue data
For each second digital signal processor unit
One digital signal input to block memory
・ In the processor unit set, performance based on red data
Arithmetic processing and other digital signal processors.
The subunit set performs arithmetic processing based on green data.
No other digital signal processor unit
In the set, arithmetic processing based on the blue data is performed.
Digital signal processor unit set
The result of the operation is transmitted to the third digital sig
Input to each of the processor units, and
Input to the digital signal processor unit
Digital signal for object identification based on the operation result
The inspection area is determined by the processor unit,
After determining the inspection area, the first digital signal processor
Input memory address by controlling the
・ The address information of the fine pixels in the inspection area from the controller
And outputs image data based on the address information.
/ D converter to convert to digital quantity and to image processing unit
Output to the image data input by the image processing unit.
Extraction and identification of color and shape features based on
Processing in the image processing unit in the system control unit.
The presence or absence of an object is identified based on the result.

[0016]

When detecting that the mustard and soy sauce are attached to a natto pack as an object, for example, the natto pack is detected as an image by the object detection unit, and the image data detected by the object detection unit is detected. Is input to the input unit, and image data input from the input unit is processed in parallel by a plurality of second processor units in a process required to extract color features and shape features by the image processing unit. The system control unit makes inferences and judgments on the image data processed in step (1) and confirms that the natto pack is accompanied by mustard and soy sauce plastic bags.

At this time, the operation of the input unit, the image processing unit, and the system control unit is realized mainly by a host computer, and a plurality of processor units are configured in a multi-layer multiplex, thereby realizing fine-grained parallel processing. Then, the target object is detected and identified at high speed.

In addition, the identification of the natto accessories described above can be performed by writing an arbitrary number of areas in a preset memory in one screen of the image detected by the object detection unit, and by setting an arbitrary number of the areas. Allocated to a plurality of parallel-configured processor units provided in the image processing unit. First, as a first step, color characteristics are applied to small areas obtained by further subdividing one area in order to accurately grasp the position of the natto pack. A large area is roughly determined by performing extraction, and a minimum area is selected from arbitrary areas required for color feature extraction and shape feature extraction later.

Next, as a second step, the area selected in the first step is colored to extract color features, or thinned to extract shape features to identify the presence or absence of an accessory.

Then, an external device such as a carry-in device for carrying the pack of natto into the input unit or a detecting unit for detecting the pack to be carried in is connected, and the external device can be controlled by the input unit. The operation for inspecting the presence or absence of the attachment of the pack can be performed quickly and reliably.

[0021]

Embodiments of the present invention will be described below with reference to the drawings.

A shown in FIG. 1 is an object device according to the present invention, C is a natto pack loading device for loading a natto pack F,
D is a defective product sorting device, which identifies the natto pack F carried in by the natto pack carry-in device by the object identifying device A, and based on the identification result, the defective product sorting device D
The soy sauce 30 and mustard 31 of the attachments shown in FIG.
If you do not have a plastic bag, eliminate the defective pack.

In FIG. 1, S is an object passage detecting section for detecting the passage of the natto pack F carried in by the natto pack carry-in device C. The position and the approximate shape of the natto pack are detected by using an optical sensor or the like. Is to be detected. Reference numeral 13 denotes a conveyor driving unit that drives the natto pack carry-in device C, and reference numeral 14 denotes a control valve that controls a defective product discharging cylinder d2, which is a sorting actuator.

As shown in FIG. 1, the object identification device A includes an object detection unit 1 for detecting a natto pack F as an object as an image, an input unit 2 for inputting image data from the object detection unit 1, An image processing unit 3 that performs processing necessary for extracting color features and shape features from image data input from the input unit 2, and a processor unit provided in the image processing unit 3 to control the image processing unit 3 And a host computer 5 for controlling the operations of the input unit 2, the image processing unit 3, and the system control unit 4, respectively.

As the object detecting section 1, a natto pack F is detected as an image using a CCD camera (solid-state imaging device camera) 32, and the detected image data is input to the input section 2.

As shown in FIG. 2, the input unit 2 is provided with a natto pack F inputted by the CCD camera 32 as the object detecting unit 1.
Is captured by the detection unit control unit 6, and the red data R, green data G, and blue data
The A / D controller 9 operates the A / D converter 9 while referring to the address information of n areas stored in the input memory address controller 7 in advance among the analog amounts of the data B. , An analog amount is converted into a digital amount, and the converted data is output to the data bus 10.

Then, the input memory address
Addresses of n areas are written in the controller 7, and the addresses are read out to the A / D controller 8, output to the address bus 11, and sent to the image processing unit 3.

U9 is a first DSP (digital signal processor) for controlling the above-mentioned operation in the input unit 2.
Unit, and its operation procedure is performed by the host computer 5.
I will be instructing from. In addition, address information of n areas stored in the input memory address controller 7 is created by a neural network assembled on the object identification DSP unit u10.

Reference numeral 12 denotes a peripheral interface unit for transmitting signals to and from an external device, such as a natto pack loading device C and a defective product sorting device D as external devices.
Of the conveyor drive unit 13, the control valve 14, and the object passage detection unit S.

As shown in FIG. 3, the image processing unit 3 comprises six second DSP units u0, u1, u2, u3, u4, u having the same internal configuration.
5 comprises a, and to perform the parallel processing by the first 2DSP unit U0～u5, the DSP unit u
0, u1, u2, u3, u4, the u5, among the image data converted into a digital amount by the A / D converter 9, once written into block memory 15 the data of the necessary n points over a data bus 10, Also, the address input from the address bus 11 is
Address information is also written to the block memory 15, and the block memo
Based on the address information and image data written in
The arithmetic processing is performed by the DSP 16 which is a pixel processor of a local parallel pipeline processing type using the plane memory 17, and the processing result is stored in the plane memory 17 and the DSP.
The data is written to the register in the register 16 and sent to the system control unit 4.

The instruction code describing the processing of the DSP 16 is loaded from the host computer 5 through the system control unit and written in the controller 18 in advance.

The other second DSP units u1 to u5 have the same configuration. Then, the second DSP unit
U0 to u5 can select any combination of single stream and multiple streams for instruction and data flow.
In particular, parallel execution of the same kind of work (a single stream of instructions and multiple streams of data) is performed by the DSP units u0 and D
A set of DSP units is composed of the SP unit u1 and
DSP unit u2 and DSP unit u3 make one set of DSP
Unit set, DSP unit u4 and DSP unit
In step u5, each job is executed as a single DSP unit set , and parallel execution of different jobs (both instructions and data are performed in a plurality of streams) is performed on the whole, and two or more operations can be performed in combination. That is, in this embodiment,
In this case, the second DSP unit u0 and the second DSP unit
In the DSP unit set with u1, the red data R
The second DSP unit u2 and the second D
Green data for the DSP unit set with SP unit u3
The arithmetic processing based on the data G is performed, and the second DSP unit u4
And the DSP unit set of the second DSP unit u5
Process by performing arithmetic processing based on blue data B
Can be distributed, and the processing speed can be increased.

The system control unit 4, as shown in FIG.
The DSP unit U6～u 11 connected by a pipeline facility interface constitutes a one-dimensional array network loop, the calculation result of the image processing section 3 a 3 D
It is sent to the object identification DSP unit u10 via the SP units u6, u7, u8, and the object identification DSP unit u10 performs inference / judgment processing on the calculation result to identify the presence or absence of the natto pack accessory. I have to. In particular, this implementation
In the example, the third DSP unit u6 has the second DSP unit u6.
DSP unit consisting of unit u0 and second DSP unit u1
Sending operation result of the set, the first 3DSP unit u7,
DS between the second DSP unit u2 and the second DSP unit u3
Sends the calculation result of the P unit set and sends it to the third DSP unit.
To u8, the second DSP unit u4 and the second DSP unit u5
And the calculation result of the DSP unit set
In addition, DSP units u6 to u11 are also locally parallel pipelines
Each processor has a
Can be implemented.

[0034] Then, when identifying the presence or absence of an accessory by the object identification DSP unit u10, the distribution of defects in natto pack from the object identification DSP unit u10 first 1 DS
The first DSP unit u9 is instructed to the P unit u9.
To the defective product sorting device D via the peripheral interface unit 12 from the
Is appropriately operated.

The DSP unit u11 controls the timing of the entire system, and the procedure of the control operation of each of the DSP units u6 to u11 is instructed from the host computer 5.

As the host computer 5, since the DSP units u0 to u11 are operated in parallel, a microcomputer is sufficient, and the cost of the object identification device A itself is reduced.

As described above, the function of each processor is the DPS
Units u0 to u5, u6 to u8, u9, u10 and u11 are divided into respective groups of the host computer 5 to perform function-distributed parallel processing. Further, the input unit 2, the image processing unit 3, and the system control unit 4 Pipeline processing is in progress.

Next, the second DSP unit of the image processing unit 3
D which is a pixel processor provided for each of u0 to u5
An allocation method to SP16 will be described with reference to FIG.

The area allocated to the six parallel DSPs 16 can be specified at random, and the size of one area l
And m can be set freely in each area, and the area is written in the memory of the controller 18.

Further, in order to increase the efficiency of the image identification processing, the processing procedure for the n regions G assigned to the six DSPs 16 of the parallel configuration is performed in the following two stages.

That is, as a first step, in order to accurately grasp the position of the natto pack, color features are extracted by colorization for a p × q small area obtained by further subdividing one area. A large area covering the entire outer shape of F is roughly determined.

At this time, if a small area g of p × q is efficiently selected at the four corners of the area G to be roughly extracted as shown in FIG.
The necessary natto pack F can be recognized, though coarse, in a seemingly wide area, and the minimum area can be determined among the areas necessary for performing color feature extraction and shape feature extraction performed later.

Next, as a second step, an arithmetic process is performed to colorize the area selected in the first step to extract color features, or to perform thinning to extract shape features, thereby obtaining the obtained color. The presence / absence of the soy sauce 30 and mustard 31 as accessories are recognized from the features and the shape features. 33 is natto.

As described above, in this embodiment, the parallel operation processing is performed by the DSP 16 having a parallel configuration of six pixels as the pixel processor. However, since the processing area is selected to the minimum necessary, the number of operations of the DSP 16 is reduced. In addition to reducing the size, the capacities of the block memory 15, the plane memory 17, etc. are also reduced, and the price of the object identification device A itself can be reduced.

Next, Table 1 shows existing image processing apparatuses and
7 shows a comparison of processing time for image data handled by one camera input.

Here, Ayako (trade name) is a typical inexpensive color image processing system, and IV-80 (trade name) is currently the fastest in a color image processing system using a microcomputer as a host computer. This is one of those that can perform image processing.

[0047]

[Table 1]

As shown in Table 1, the object identification device A according to the present embodiment, the aforementioned Ayako and IV-80 use 32 (l).
As a result of measuring the processing time for gradation conversion, first-order differentiation, binarization, and K median filter for × 32 (m) × 32 (n) = 32,768 pixels,
It can be seen that the processing of the object identification device A is about seven times faster than that of the IV-80, and that the processing is so fast that it cannot be compared with Ayako, that is, 100 times or more.

As described above, the object discriminating apparatus A according to the present embodiment can perform a high-speed processing seven times or more as compared with the conventional image processing system as described above on the hardware side, and also can perform the processing on the software side. As described above, the color feature and the shape feature are extracted in two stages, and the identification of the object such as the natto pack and the cap of the beer bottle can be performed about 10 times as fast as that of the hard surface. When the software surfaces are combined, the object can be accurately identified by a high-speed processing that is nearly 100 times faster than before.

In this embodiment, the object identification device A is capable of identifying at a high speed whether or not there is an accessory of the natto pack F.
However, it is natural that the present invention can be applied to other cases, such as detection of caps of beer bottles, identification of parcels and postal items for courier service, reading of addresses, seafood such as seashells, shellfish, etc., etc. It is also possible to efficiently sort agricultural products and identify defective products such as coins and pottery.

[0051]

According to the present invention, an input unit, an image processing unit,
And each digital signal provided in the system controller
Each processor unit has local parallel pipeline processing.
Processors that perform
Operation in parallel with a computer, and the input
The output image data is analog data of red data and green data.
Color data and blue data are stored in the detection unit control unit.
And the first digital signal processor unit
Input memory address based on the control by
To the coarse pixel address information output from the controller
Based on the specified address information, red data and green data
Digital data and blue data by A / D converter
Output to the image processing unit, and the image processing unit
Digital data of red data, green data, blue color
The data is transferred to each second digital signal processor unit.
Input to the block memory of the
The signal processor unit set is based on the red data.
And perform other digital signal processing.
The processing unit based on green data is
Processing and other digital signal processor units.
The knit set performs arithmetic processing based on blue data.
Digital signal processor unit set
The result of the calculation in the
Input to each signal processor unit,
Enter the third digital signal processor unit
Digital system for object identification based on the
Inspection area judgment by the signal processor unit
The first digital signal
・ Input memory
A fine pixel of the inspection area from the address controller
Output dress information, and image data based on the address information.
Data is converted to digital quantity by A / D converter
Image output to the processing unit and input in the same image processing unit
Extraction and identification of color and shape features based on data
Process, and in the image processing section in the system control section.
The presence or absence of an object based on the processing result
To improve the accuracy and speed of identification of required objects
And a small, inexpensive object identification system as a whole.
Can be provided.

[Brief description of the drawings]

FIG. 1 is a conceptual explanatory view of an object identification device according to the present invention.

FIG. 2 is a block diagram of an input unit.

FIG. 3 is a block diagram of an image processing unit.

FIG. 4 is a block diagram of a system control unit.

FIG. 5 is an explanatory diagram of an assignment method of a pixel processor.

FIG. 6 is an explanatory diagram of feature extraction of a natto pack.

[Description of Signs] A Object identification device 1 Object detection unit 2 Input unit 3 Image processing unit 4 System control unit 5 Host computer

──────────────────────────────────────────────────続 き Continued on front page (58) Field surveyed (Int.Cl. ⁷ , DB name) H04N 7/18

Claims (1)

(57) [Claims] An object conveyed by a conveying means is a solid.
Object detection that detects with an image sensor camera and outputs image data
Out part (1), the image data output from the object detection unit (1) Digitally
The input unit (2) for image conversion and the arithmetic processing of image data input from the input unit (2) are performed.
Image processing unit (3), and image data processed by the image processing unit (3).
A system control unit (4) that performs another processing, the input unit (2), the image processing unit (3), and a system control unit
The host computer (5) that controls the operation of (4) and
An input unit (2) is provided with an image data output from an object detection unit (1).
The detector control unit (6) that captures the
・ Based on control from signal processor unit (u9)
Input that outputs the address information of the image data
Memory address controller (7) and same input
・ Output from memory address controller (7)
Based on the address information, it is sent to the detector control unit (6).
The corresponding address information data of the loaded image data.
Equipped with A / D converter (9) for converting to digital quantity
The image processing unit (3) receives the digital data input from the input unit (2).
A block memory (15) for temporarily writing the amount of image data,
Based on the image data written in the block memory (15)
Digital signal processor
(16) and the digital signal processor (16)
The plane memory (17) used for computation and the host
Digital signal based on computer (5) control
-Consists of a controller (18) that controls the processor (16)
Second digital signal processor unit formed
Are configured as a set of at least two
Multiple signal processor unit sets
The system control unit (4) controls the digital processing of the image processing unit (3).
Connect to each signal processor unit set
A third digital signal processor Uni Tsu bets were,
One with the third digital signal processor unit
Connected by a 1D digital network
Object to control the signal processor unit (u9)
Separate digital signal processor unit (u10)
The input unit (2), the image processing unit (3), and the system control unit (4)
Each digital signal processor unit provided
Processors that perform local parallel pipeline processing
And they are connected in parallel by the host computer (5).
Activate the input unit (2) and input the image data output from the object detection unit (1).
Red data (R), green data (G), blue
The data is taken into the detection unit control unit (6) as data (B) and
By one digital signal processor unit (u9)
Input memory address con
The coarse pixel address information output from the controller (7)
Based on predetermined address information red data (R), green
A / D converter for data (G) and blue data (B) respectively
In (9), it is converted to digital quantity and output, and in the image processing unit (3), the digital data input from the input unit (2)
Red data (R), green data (G), blue data (B)
For each second digital signal processor unit
Input to the block memory (15), one digital signal
Red data (R) in null processor unit set
Operation based on other digital signals
Based on green data (G) in processor unit set
Performs arithmetic processing and performs other digital signal processing.
Processing unit based on the blue data (B)
Process for each digital signal processor unit set.
Calculation result in the third digit of the system control unit (4)
Input to each signal processor unit,
Enter the third digital signal processor unit
Digital system for object identification based on the
The inspection area is controlled by the signal processing unit (u10).
Judgment is made, and after judging the inspection area, the first digital signal processor
Input memory address by controlling the
Address of fine pixels in the inspection area from the
Output dress information, and image data based on the address information.
Converts over data to the digital quantity by the A / D converter (9)
Output to the image processing unit (3) and input in the same image processing unit (3)
Of color and shape features based on extracted image data
・ Perform identification processing, and
The presence or absence of an object based on the processing result in the image processing unit (3)
An object identification device that performs identification.