JPH0797405B2 - Image recognition device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION [Industrial field of application] The present invention relates to patterns used for position detection, visual inspection, etc.
The present invention relates to an image recognition device using a matching method.

[Outline of Invention]

The present invention is a pattern used for position detection, visual inspection, etc.
Regarding an image recognition device using the matching method, image data is temporarily stored in an image memory from a camera, and a part of the image data and a part of the reference image data stored in another image memory are EX- Even if these image data are misaligned by using an OR operation circuit to make a hardware comparison, image recognition that can quickly determine the amount of misalignment by processing multiple times at high speed The device can be provided.

[Conventional technology]

The conventional image recognition device that uses the pattern matching method used for position detection, visual inspection, etc., uses a standard pattern with a size of 3 × 3 to 30 × 20 over the entire screen when performing high-speed processing with hardware. There was pattern matching locally.

[Problems to be solved by the invention]

However, in the above-mentioned conventional technology, in order to perform high-speed processing over the entire screen, EX-OR elements and the elements of the output addition circuit are required for the number of pixels matching the pattern matching circuit, and the matching screen is large. Then, there is a problem that the hardware configuration becomes extremely complicated.

When a conventional image recognition device is used, it is necessary to attach a special mark having a characteristic so that it can be found in a minute area, or to provide a characteristic portion as a substitute for the mark, in order to detect the position in a minute area. In addition, because the matching screen cannot be made large, the visual inspection was not an inspection.

Therefore, the present invention solves such problems,
Its purpose is to process pattern matching between a reference image and an image to be recognized in a large area at high speed, and to pattern match two images with an arbitrary shift amount. Thereby, the position is detected and the appearance is inspected.

[Means for solving problems]

The image recognition apparatus of the present invention includes a first image memory that stores reference image data, a second image memory that stores image data obtained through a camera, and an arbitrary image memory in the first image memory. A first address generating means for generating a first address indicating an area and a second address generating means for generating a second address shifted with respect to the first address; The data of the first address and the data of the second address of the second image memory are input to an arithmetic circuit that calculates an exclusive OR (EX-OR) and the sum is calculated, and is calculated using this arithmetic circuit. It is characterized in that the amount of shift is changed until the sum becomes minimum, and the calculation is repeated.

[Action]

If the image data input to the image recognition apparatus of the present invention through the camera has a deviation from the reference image data, it is necessary to repeat the EX-OR operation a plurality of times in order to determine the deviation amount. That is, if the shape and position of the reference object and the object to be recognized are substantially the same, the EX-OR of the data corresponding to each pixel of the first image memory and the second image memory is taken. The value becomes 0, and the sum becomes extremely 0 even if added over a wide area. However, if there is a slight positional deviation in the object to be recognized, the image data input via the camera is displayed horizontally or Within the input image data, shift in either vertical direction and compare with the reference image data, and find the shift position that minimizes the area of the part that does not match (that is, the sum of EX-OR). The object to be recognized can be relatively moved to the same place in the image data as the reference, and the amount of movement due to the shift at that time can be obtained as the amount of deviation.

The image recognition device of the present invention is an EX-
The time required to compare two image data is shortened by processing the operation of calculating the sum of ORs in a hardware manner using an arithmetic circuit. Further, by temporarily storing the image data from the camera in the image memory, the time for inputting the image from the camera each time the calculation is performed is omitted, and the calculation process can be repeated only by generating the shifted address. Furthermore, by storing the image data from the camera in the image memory, rather than comparing all the image data obtained from the camera with the reference image data, an arbitrary part of the image data is extracted and used as the reference. Compared with the image data, the amount of data to be processed is reduced.

As described above, the image recognition apparatus of the present invention can perform comparison processing of image data at an extremely high speed by adopting the above-described configuration, and can detect the deviation amount between the image data input through the camera and the reference image data. It is possible to make a quick decision.

Furthermore, since the address to be compared in each image memory can be set arbitrarily, several methods for obtaining the deviation amount can be adopted. For example, in the case where the shift amount becomes large, it is possible to quickly and accurately obtain the shift amount by setting the address shift amount in increments of ± N instead of ± 1.

〔Example〕

FIG. 1 is a block diagram of an embodiment of the present invention. The procedure will be described below.

First, an object to be recognized is placed at a reference position. The input image from the television camera 101 is input to the reference image memory 103b according to the signal of the address generator 109 synchronized with it, and this is used as reference image data, and the preparation is completed.

Next, an image of an object for which position detection / appearance inspection is desired is input from the television camera 101 to the work image memory 103a in accordance with a signal from the address generator 109 synchronized with the image.

Next, the address generator 109 generates an address signal of an area having an appropriate size in the image memory, and the output signal 106b of the reference image memory 103b and the output signal 106a of the work image memory-103a are input to the EX-OR110. To do. The outputs of the EX-OR 110 are added by the adder 111 according to the signal from the address generator 109. The addition data obtained here is the total amount of difference for each pixel of the two image memories.

Horizontal address 107a and vertical address 107b are address generators
Although it is generated from 109, there are adders 104a and 104b between the horizontal address 105a and the vertical address 105b to the work image memory 103a, and the adders 104a and 104b are added or subtracted depending on the horizontal and vertical shift amounts of the address generator 109. It Then, the reference image memory 103b and the work image memory 10
3a is relatively shifted by the shift amount set in the horizontal and vertical directions.

In this example, the work image memory is provided with an adder, but the same effect can be obtained even if the reference image memory is provided with an adder.

FIG. 2 is a diagram for explaining movement of image data. The reference image data and the work image data are shown at the same time to show the positional relationship between each other.

FIG. 2a shows an image 203b captured as a reference image in the reference image memory 103b. The entire image memory is 201, but the area required as a reference image may be a fraction thereof, and here is the area indicated by 202.

FIG. 2B shows a state in which the target object 203a to be recognized is captured in the work image memory, and is usually displaced from the reference image 203b because there is a position shift. The addition means of the present invention represents the number of pixels in which the two images are misaligned.

FIG. 2c shows a state in which the shift amount is appropriately set in the horizontal direction and the straight direction. If the shift amount is set properly, the number of misaligned pixels will decrease, and if the setting is incorrect, the number of misaligned pixels will increase. In the present invention, the shift amount is controlled and increased or decreased so that the value of the adding means becomes as small as possible.

FIG. 2d shows a state in which the shift amount is optimized. The number of misaligned pixels does not become 0 because of a digital error, but is the minimum value. The shift amount at this time is the amount of positional deviation or a proportional amount thereof.

There are various methods for obtaining the optimum value of the shift amount, but here, two methods are shown. One is to find the number of shift pixels in four directions (horizontal ± 1 shift, vertical ± 1 shift) with respect to the current shift amount, and set a shift position indicating a smaller addition amount as a new shift amount. If all four peripheral directions show a value larger than the current displacement pixel number, it is the minimum (minimum), so it is determined that the displacement amount has been found. Secondly, in the above method, in order to avoid the drawback that it is easy to make a mistake with a larger minimum value before reaching the optimum value, the number of shift pixels for each fixed amount of shift in the horizontal and vertical directions is spread over a wide range in advance. This is a method of taking a matrix and starting from the position of the shift amount where the value is the minimum, and searching for the optimum value by the above method.

Furthermore, instead of the above four directions, a method of alternately searching every two directions,
There is a method to search for ± N shifts instead of ± 1 shifts.

The processing speed in one EX-OR / addition processing according to the present invention is as follows assuming that a non-interlaced camera (about 17 milliseconds for capturing one screen) is used. If you use the same clock speed as the camera when you EX-OR the whole screen, it takes about 17 milliseconds, which is almost the same as the screen capture. If the size of the reference image is a quarter of the screen, the processing speed will be about 4 milliseconds. This process can be executed in about 0.4 seconds when the shift amount is found 100 times.
Furthermore, higher-speed processing can be expected by increasing the speed of the clock and reducing the area of the reference image.

It is possible to have a circuit configuration that has only a standard screen that can be shifted, and directly executes EX-OR / addition without loading the image from the TV camera into the memory, but one EX-OR /
It always takes about 17 milliseconds for the addition, which is not a practical speed, and it is impossible to devise a further speedup.

〔The invention's effect〕

As described above, according to the present invention, the position of an object to be recognized can be detected at high speed with a simple circuit, and the shape of an object placed at an arbitrary position can be inspected by a pattern matching method. Have an effect.

[Brief description of drawings]

FIG. 1 is a block diagram showing a position embodiment of an image recognition apparatus of the present invention. FIG. 2 is a movement explanatory view of the image data showing the relationship of each image to be EX-ORed. FIG. 2a is a diagram in which the reference image is read. Fig. 2b shows the positional relationship with the reference image when the image to be recognized is read, and Fig. 2c shows the optimum shift amount during the search for the position, and the two images match. The figure which shows time. 101 …… TV camera 102 …… Input signal from TV camera to image memory 103a …… Work image memory 103b …… Reference image memory 104a, b …… Adder (addition / subtraction circuit) 105a, b …… Image memory Address input of 103a 106a, b ... Output signal from image memory 103a, b 107a, b ... Address input of image memory 103b 108 ... Horizontal / vertical shift amount 109 ... Address generator 110 ... EX-OR 111 …… Adder 201 …… Total area of image memory 202 …… Reference image area 203a …… Image of recognition object (work) 203b …… Reference image

Claims (1)

[Claims] 1. A first image memory for storing reference image data, a second image memory for storing image data obtained through a camera, and an arbitrary area in the first image memory. First address generating means for generating a first address shown, second address generating means for generating a second address shifted with respect to the first address, and the first address of the first image memory. Data of the address and the data of the second address of the second image memory are input, and an arithmetic circuit that calculates an exclusive OR of the data and calculates a sum is calculated. The data of the first address of the first image memory and the data of the second address of the second image memory are supplied to the arithmetic circuit by changing the shift amount until An image recognition device characterized by repeating.