JP3483039B2 - Identification device based on image signals - Google Patents

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image signal identification apparatus capable of identifying an object by a CCD camera at high speed by processing an image signal obtained by a CCD camera or the like.

[0002]

2. Description of the Related Art Conventionally, there is a method using a black and white image processing technique for selecting fish species from the shape of the fish body.

That is, such an apparatus illuminates a fish body flowing on a semitransparent belt conveyor from below the semitransparent belt conveyor, detects a silhouette fish body with a black and white camera, and recognizes it by a grayscale image. It is something to do.

In such a measurement process, fish bodies are distinguished by an image processing technique of density conversion, and an attempt is made to select fish species by obtaining contours of fish bodies and measuring body height, body length and the like.

[0005]

However, in the device for recognizing with a grayscale image as described above, since the fish of the same family have similar shapes, sufficient recognition rate cannot be obtained only by the shape information. , Not practical. Therefore, it has not been put to practical use.

In order to improve the recognition rate, it seems effective to extract both the shape feature and the color feature using a color image taken by a color camera, instead of using a black and white image of a black and white camera.

However, if the conventional color image processing apparatus is used, the same color image processing apparatus targets all the data of one screen, so that the minimum necessary information of the fish body is taken in and processed at high speed. Therefore, there is a problem that a large-capacity memory is required, wasteful access takes time, and fish species cannot be identified at high speed.

[0008]

Therefore, in the present invention, a detector for outputting an image signal, an image signal distributor for extracting and outputting an image signal at a predetermined position from the image signal output from the detector, An understanding unit that outputs the result of identification performed using the image signal at the predetermined position output from the image signal distribution unit to the outside, and a host computer that controls and monitors the operations of the detection unit, the image signal distribution unit, and the understanding unit. In the identification device based on the image signal, the image signal distribution unit first divides the image signal for one screen output from the detection unit into a preset mesh shape, and divides each divided mesh area. Only the image signal at the predetermined position in is input to the understanding unit, and the understanding unit discriminates from the background image based on the input image signal to mesh the existence position in one screen. The image signal distribution unit inputs the image signal of the small region specified by the understanding unit to the understanding unit, and identifies the small region composed of one or a plurality of mesh-shaped regions. Based on the input image signal, the understanding unit performs shape extraction by performing identification with the background image, and provides an identification device based on the image signal configured to perform identification based on the extracted shape information. Is.

Further, the understanding unit specifies a color feature extraction area for color feature extraction based on the shape information extracted by the shape extraction, and a histogram of density values based on the image signal in the specified color feature extraction area. It is also characterized in that it is configured to perform discrimination based on color feature extraction by analyzing the distribution of.

[0010]

When the identification device based on the image signal of the present invention is used as a fish species identification device, for example, a large amount of fish caught in a fishing boat and iced is melted in an aquarium or the like,
Separate the fish into individual fish, set a predetermined interval, and remove water from the surface of the fish with a drainer to prevent reflection from the water film on the surface of the fish. The fish bodies that have been separated and are conveyed at predetermined intervals are targeted for fish species identification.

In the detecting section, the fish body is photographed by a color CCD camera and sent to the image signal distributing section as an analog image signal.

In step 1, the image signal distribution unit divides the image signal for one screen into a preset mesh shape, and converts only the image signal at a predetermined position in each divided mesh area into a digital image signal. Then, in step 2, as will be described later in step 1, only the image signal of the small area specified by the understanding part is converted into a digital image signal and input into the understanding part.

In step 1, the understanding unit recognizes the existing position in one screen on a mesh-like region basis by discriminating the subject from the background image of the color CCD camera based on the input image signal. Specify a small area consisting of one or more mesh-shaped areas, and
Then, from the above result, image information of a plurality of regions for measuring the shape in more detail and a region for color feature extraction is input from the image signal distribution unit and assigned to the image-dedicated processors of the plurality of processor units of the understanding unit. The image dedicated processor also uses an image memory. Then, a plurality of image-only processors perform detailed shape feature extraction under the control and support of the general-purpose processor of each processor unit, and if the information can be identified only by the shape features, the result is sent to the fish species distribution device, If the information cannot be identified using only the shape feature information, the size and position of the area for color feature extraction are determined based on the shape feature information, and color averaging, histogram, co-occurrence are performed by multiple processor units. Matrix, FFT processing, HSV conversion, and other processing are performed on the color feature extraction area, the shape and color feature extraction information are comprehensively determined, the fish body is identified in detail, and the fish species and size are identified. .

The above operation is performed by the host computer section which controls and monitors by integrating the high-speed communication between the general-purpose processors constituting the image signal distribution section and the understanding section and the control section and the detection section and the image signal distribution section. The data can be processed at high speed by the pipeline operation, and the fish can be identified quickly and reliably.

In the image signal identification device of the present invention, the object is identified by the CCD camera as described above. If the identification device is a fish identification device, the identification result is automatically detected, for example. Can be sent to a fish species distribution device or the like configured to distribute the fish, and the fish can be sorted according to fish species and size, and can be boxed in fixed quantities.

[0016]

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

FIG. 1 shows a case where the image signal identification device according to the present invention is used as a fish species identification device K.
A is a transportation unit that arranges unsorted fish bodies one by one at a predetermined interval and transports them while removing water on the surface of the fish. K is a fish body understanding device that identifies the fish species and size.
E is a fish species distribution device that separates fish bodies into boxes according to the identification result obtained from the understanding part D, and F is the transport part A, the image signal distribution part C, the understanding part D, and the fish species distribution device. A host computer unit that controls and monitors the operation of E.

The transporting section A melts the ice of the thrown fish bodies and circulates the water between the water tanks 4 for separating the fish bodies one by one to drain the fish bodies one by one. A water flow carrier 5 for transporting the fish, a drainer 3 for temporarily removing water on the surface of the fish with a sponge and cold air, and a transporter for transporting the drained fish one by one under the color CCD camera 2 of the detection unit B. It is composed of 7 and.

As shown in FIG. 2, the water tank 4 is provided with a separating device 4a for separating fish bodies one by one. The separating device 4a includes an arm 4c that automatically expands and contracts on a rotating shaft 4b that rotates at a constant speed by a motor drive.
Are arranged in a radial manner, and a fish placing plate 4d is connected to the tip of the same arm 4c. The fish in the aquarium 4 is scooped up by the fish placing plate 4d one by one by the rotation of the separating device 4a. When reaches a position higher than the start end of the water flow carrier 5,
The arm 4c is extended, and then the arm 4c is bent downward so that the fish body loading tray 4d faces downward, so that the fish bodies are carried one by one to the water stream carrying device 5.

It should be noted that the number of fish bodies to be carried to the water stream carrying device 5 can be increased by the number of separating devices 4a installed in the water tank 4.

By adjusting the shape and size of the fish loading plate 4d, the fish can be scooped up one by one with certainty.

Since the water stream is mainly used to carry the fish in the carrying section A having such a structure, it is possible to prevent the fish from being damaged.

As shown in FIG. 3, the draining device 3 is composed of a net-like or slat-like water-permeable conveyor 3a, a motor-driven sponge 6a that moves up and down, and a cold air blower 6b. The water on the surface of the fish, which is conveyed one by one by 3a, is removed, the remaining water on the surface is absorbed by the sponge body 6a, and the water on the surface of the fish is removed by the cool air from the cool air blower 6b. It prevents the trouble of shooting due to the reflection of.

The fish comprehension apparatus K comprises a detection section B, an image signal distribution section C, an understanding section D, and a host computer section F for controlling them.

As shown in FIG. 1, the detection section B is composed of an illumination lamp 1 arranged above the carrying device 7 and a color CCD camera 2 with an electronic shutter.
The color CCD camera 2 captures an image of the fish on the transporting device 7 exposed to the above light to obtain an analog RGB image signal, and the analog RGB image signal is input to the image signal distribution unit C.

In the image signal distributing section C, as shown in FIG. 4, analog RGB input from the color CCD camera 2 is input.
The image signal is converted into a digital image signal, and the signal is distributed and input to the first processor unit 14-1, the second processor unit 14-2, and the third processor unit 14-3 of the understanding unit D described later. It is composed of an A / D conversion / image signal distribution board 11 and an image signal distribution control DSP board 12 for controlling the operation of the board 11.

As shown in FIG. 4, the understanding section D includes a first processor unit 14-1 and a second processor unit 1
It is composed of 4-2 and a third processor unit 14-3. Each processor unit 14-1, 14-2, 1
As shown in FIG. 5, 4-3 includes a plurality of image-dedicated processor boards 15-1 to 15-n, a plain memory board 16, and a general-purpose processor board 17, respectively. The image-dedicated processor boards 15-1 to 15-n are
Each is composed of a first block memory 15-a, a second block memory 15-b, a matrix gate 15-c, and an image dedicated processor 15-d.

Then, the image signal distribution unit C and the understanding unit D
The constituent elements of are connected by the image signal bus 13.

In the image signal distributor C, among the analog RGB image signals input from the detector B, in step 1,
As shown in FIG. 6, the image signal for one screen is divided into preset meshes, and only the image signals at predetermined positions in the divided mesh-like regions are converted into digital image signals to obtain the image signal bus 13. Through the image signal bus 13 to the understanding section D, and in step 2, only the image signal of the small area G specified by the understanding section D is converted into a digital image signal in step 2, and the understanding section D is inputted through the image signal bus 13.
First processor unit 1 provided in the understanding section D
4-1, the second processor unit 14-2, and the third processor unit 14-3 are distributed and processed.

Each processor unit 14-1, 14-
When inputting a digital image signal to the second and 14-3, the first block memory 15 is processed by pipeline processing.
-A and the second block memory 15-b, the digital image signal is stored by inputting it to either the first block memory 15-a or the second block memory 15-b which is not being processed. We are trying to speed up the transmission. Then, the digital image signal stored in the first block memory 15-a or the second block memory 15-b is used to perform the calculation by the image-dedicated processor 15-d, and the calculation result is input to the general-purpose processor board 17.

The host computer section F realizes high-speed processing by executing parallel processing of the plurality of image-dedicated processor boards 15-1 to 15-n and the plurality of plane memory boards 16-1 to 16-n. ing.

Using these results, the general-purpose processor board 17 performs high-speed arithmetic processing such as FFT processing and identification, and further, the first processor unit 14-1.
By parallelly processing the second processor unit 14-2 and the third processor unit 14-3, it is possible to perform shape feature and color feature extraction and fish species identification at high speed.

In this way, the image distributing unit C plays a role of speeding up the processing by compressing the data amount by selecting only the necessary digital image signal in one screen, and the parallel processing is performed to perform the operation speed. Further, the understanding unit D is responsible for the speedup of identification, and the speedup by pipeline processing of the entire system is performed by the two block memories 15-a, 1
5-b and the host computer unit F play a role in realizing the speedup of fish species identification by a color image by their synergistic effect.

Next, the above-mentioned area setting will be described.

FIG. 6 and FIG. 7 are explanatory views of a method of setting a small area G for performing edge extraction for measuring the total length and the body height as an example of shape feature extraction. As shown in FIG. 6, the original image of horizontal X × vertical Y pixels is divided into a preset mesh shape,
Only the image signal of a pixel at a predetermined position in each divided mesh region is extracted, and the understanding unit D performs the difference between the Kirsch operator or the color average of the fish and the background image with respect to the three primary colors, and in one screen. The existence position is recognized for each mesh-shaped area, and the small area G including one or a plurality of mesh-shaped areas is specified.

Next, in step 2, the image signal distribution unit C inputs the image signal of the small region G to the understanding unit D based on the small region G specified by the understanding unit D in the above step 1, and the understanding unit D receives the image signal. At D, shape extraction is performed by performing discrimination from the background image based on the input image signal, and discrimination is performed based on the extracted shape information.

In particular, in the case of identifying a fish species, it is possible to first measure the total length and height of the identified fish by using the extracted shape information.

Then, based on the measured total length information and body height information, the area H for color feature extraction is specified.

Next, fish species identification by color feature extraction will be described.

As an example, sesame mackerel and chub mackerel of the same family are targets for fish species identification, and as shown in FIG. 9, the color feature extraction area H is set in the abdomen of the fish body where the color features often appear. As described above, the total length and height of the fish body are measured within an error range of ± 10% or less, so when setting the position of the area for color extraction, it is possible to extract color features even if an error of ± 10% occurs. It is sufficient to secure the size of a large area.

FIG. 10 is a density distribution in the entire length direction of sesame mackerel, which is scanned on the broken line x in FIG. 9 from the head through the color feature extraction region H of the abdomen in the caudal fin direction. Since it is stable, it can be seen that the shift in the size and position of the color feature extraction area H in the entire length direction has little influence on the color feature extraction result. It turns out that feature extraction is possible.

11 to 13 show the density value of sesame mackerel when the size of the color feature extraction area H is changed and the set position in FIG. 9 is changed in the vertical direction by ± 10 pixels (about ± 10% of the body height). A histogram is shown.

The size of the color feature extraction region H is 32 pixels in the horizontal direction and 4 pixels in the vertical direction in FIG. 11, 32 pixels in the horizontal direction and 24 pixels in the vertical direction in FIG. 12, and 32 pixels in the horizontal direction and 32 pixels in the vertical direction in FIG.
When the color feature extraction region H becomes smaller as shown in FIG. 12, the distribution of the histogram of the density values changes largely with respect to the fluctuation of ± 10 pixels, but as shown in FIG. 12, the size becomes about 32 pixels horizontally and 24 pixels vertically. Then, the error of the position setting does not appear, the change of the average value of the histogram of the density value is 71 to 79, and the change of the variance is 77 to 86, and the density value histogram (FIG. 14) of the mackerel extracted in the same manner. Change in average value 91 to 9
4. Since the change in dispersion is 30 to 53, both values can be sufficiently separated.

Therefore, by analyzing the distribution of the density value histogram in the color feature extraction area H determined by such a procedure, it is possible to sufficiently identify the fish species, and the color feature extraction area H is small. Including the shape feature extraction, it is possible to identify fish species at high speed.

To identify the fish species, the color feature extraction area H
It is sufficient to set the size of 32 pixels in the horizontal direction and 24 pixels in the vertical direction (FIG. 12) to 32 pixels in the horizontal direction and 32 pixels in the vertical direction (see FIG. 1).
3) It can be seen that the size is not so large.

Further, by setting a plurality of the color feature extraction areas H, the identification accuracy can be improved.

The fish species distribution device E includes a water flow carrier device 8, a water flow distribution / sorting device 9, and a plurality of sorting boxes 9d-1 to 9d.
-N and.

The water stream carrying device 8 carries the fish bodies from the carrying device 7 to the water stream distributing / sorting device 9 using the water stream.

As shown in FIG. 15, the water flow distribution / sorting device 9 includes a central water channel 9a and a central water channel 9a attached in the transport direction.
a plurality of branch water channels 9b-1 to 9b-n branched from a,
It is composed of water jet outlets 9c-1 to 9c-n provided at the start ends of the respective branch water channels 9b-1 to 9b-n, respectively, and at the end of each of the branch water channels 9b-1 to 9b-n. Are provided with sorting boxes 9d-1 to 9d-n containing ice, and the fish bodies received from the water stream carrier 8 are the water jets selected based on the fish species identification information obtained in the understanding section D. Divergence water channels 9b-1 to 9b-n depending on the type of fish depending on the water ejected from the mouth
And sort boxes 9d-1 to 9d-n for the relevant fish species.
Separately collect.

[0051]

According to the first aspect of the present invention, there is provided a detecting section for outputting an image signal, and an image signal distributing section for extracting and outputting an image signal at a predetermined position from the image signal output from the detecting section. , An operation of the understanding unit that outputs the result of the identification performed using the image signal at the predetermined position output from the image signal distribution unit to the outside, and the operation of the detection unit, the image signal distribution unit, and the understanding unit.
In the image signal identifying apparatus including a host computer unit for monitoring, first, the image signal distribution unit divides the image signal for one screen output from the detection unit into a preset mesh shape, and divides the divided mesh. In the understanding unit, only the image signal at a predetermined position in each region of the shape is input to the understanding unit, and the understanding unit discriminates the background image based on the input image signal to determine the existence position in one screen in the mesh region. Recognizing in units, a small area consisting of one or a plurality of mesh areas is specified, and then the image signal distribution unit inputs the image signal of the small area specified by the understanding unit to the understanding unit, and the understanding unit Then, based on the input image signal, shape extraction is performed by distinguishing from the background image, and it is configured to perform identification based on the extracted shape information. , Very small amount of data to be processed, the image processing faster, it is possible to speed up the identification process.

According to the second aspect of the present invention, the color feature extraction area for color feature extraction is specified based on the shape information extracted by the shape extraction, and based on the image signal in the specified color feature extraction area, By performing the classification based on the color feature extraction by analyzing the distribution of the histogram of the density values, the reliable classification process can be performed.

[Brief description of drawings]

FIG. 1 is an explanatory diagram showing a configuration of a fish species identification device according to the present invention.

FIG. 2 is a side view of a water tank and a water stream carrying device of a carrying section.

FIG. 3 is a side view of the water draining device of the transport section.

FIG. 4 is an explanatory diagram showing a configuration of an image signal distribution unit and a fish body understanding unit.

FIG. 5 is an explanatory diagram showing a configuration of a fish body understanding unit.

FIG. 6 is an explanatory diagram of a small area setting method.

FIG. 7 is an explanatory diagram of a small area setting method.

FIG. 8 is a graph showing the relationship between the number of pixels in a small area and the block memory capacity.

FIG. 9 is an explanatory diagram showing setting positions and sizes of color feature extraction areas.

FIG. 10 is a density value histogram in the body length direction.

FIG. 11 is a density value histogram of a color feature extraction area (Sesame mackerel).

FIG. 12 is a density value histogram of a color feature extraction area (Sesame mackerel).

FIG. 13 is a density value histogram of a color feature extraction area (Sesame mackerel).

FIG. 14 is a density value histogram (masaba) of the color feature extraction area.

FIG. 15 is a plan view of a water flow distribution / sorting device.

[Explanation of symbols]

B detector C image signal distributor D Understanding Department F Host computer section G small area H color feature extraction area 2 color CCD camera 14-1 First processor unit 14-2 Second processor unit 14-3 Third processor unit 17 General-purpose processor board 15-a First block memory 15-b Second block memory 15-c Matrix gate 15-d Image dedicated processor 15-1 to 15-n Image-only processor board 16-1 to 16-n plain memory board

─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Sumio Nagata 3-6-204 Miyakoji, Omura-shi, Nagasaki Prefecture (72) Inventor Shikata 3-3-302 Hirota, Sasebo City, Nagasaki Prefecture (56) References JP-A-6-6793 (JP, A) JP-A-5-159064 (JP, A) JP-A-5-20455 (JP, A) (58) Fields investigated (Int.Cl. ⁷ , DB name) G06T 7/00-7/60 G06T 1/00 H04N 7/18

Claims (2)

(57) [Claims] 1. A detector (B) for outputting an image signal, and an image signal distributor (C) for extracting and outputting an image signal at a predetermined position from the image signal output from the detector (B). The understanding unit (D) that outputs the result of the identification performed using the image signal at the predetermined position output from the image signal distribution unit (C), the detection unit (B), and the image signal distribution unit (C). In the identification device based on an image signal, which includes a host computer unit (F) that controls and monitors the operation of the understanding unit (D), first, the image signal distribution unit (C) is output from the detection unit (B). The image signal for one screen is divided into preset mesh shapes, and only the image signal at a predetermined position in each divided mesh area is input to the understanding unit (D).
The understanding unit (D) recognizes the existing position in one screen in mesh area units by discriminating it from the background image based on the input image signal, and recognizes it from one or a plurality of mesh areas. Then, the image signal distribution unit (C) inputs the image signal of the small region (G) specified by the understanding unit (D) to the understanding unit (D).
The understanding unit (D) is configured to perform shape extraction by performing discrimination from a background image based on the input image signal, and to perform discrimination based on the extracted shape information. Identification device. 2. The understanding unit (D) specifies a color feature extraction area (H) for color feature extraction based on the shape information extracted by the shape extraction, and an image in the specified color feature extraction area (H). The image signal identifying apparatus according to claim 1, wherein the image signal identifying apparatus is configured to perform color feature extraction-based identification by analyzing a distribution of a density value histogram based on the signal.