JPH10312460A - Image processing method and high-precision image processing device - Google Patents

Abstract

(57) Abstract: In an image processing apparatus that detects a straight line from an image using Hough transform, a quantization error generated when performing an Hough transform on an image in an area having a small XY value is reduced. SOLUTION: X out of images taken into an image memory is provided.
An image of a region having a small Y value is extracted, the extracted image is enlarged and stored in an enlarged image memory having substantially the same number of pixels as the original image memory, and each candidate point of the image stored in the enlarged image memory is Huff transform.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image processing method and a high-precision image processing apparatus for detecting a straight line portion from various images and detecting the inclination of the straight line with high accuracy.

[0002]

2. Description of the Related Art Conventionally, as one of image processing techniques, there is a straight line detection technique for detecting a straight line portion from an image picked up by an imaging device. According to this straight line detection technology, for example, it is possible to detect a flaw or the like attached to a product, or to detect a boundary between a wall and a floor, and to guide the boundary to drive an automatic traveling vehicle. Various applications are considered.

FIG. 4 shows an example of a conventional image processing apparatus.
In the figure, reference numeral 10 denotes an image processing device, and reference numeral 20 denotes an imaging device. The image processing device 10 is generally configured by a computer. As is well known, the computer is a central processing unit 11, a read-only memory 12, a rewritable memory 13 for temporarily storing a program or input data, an input port 14, an output port 15, and the like. Composed of

The read-only memory 12 stores information such as characters to be displayed on a screen. The rewritable memory 13 has an area used as an image memory 13A for capturing an image captured by the imaging device 20, and an area for storing curve data on the θ-ρ plane generated at the time of Hough transform described below (hereinafter referred to as a curve). In addition to providing a group memory 13B, the central processing unit 11 is
A program operating as C, a program operating as a binarizing unit 13D for normalizing pixels of an image into binary pixels of white or black, a program operating as a Hough transforming unit 13E, and a program operating as a frequency distribution searching unit 13F , A program to be operated as the straight line extracting means 13G is stored. These programs are transferred from an external storage means such as a floppy disk drive to the rewritable memory 13 and stored therein at the time of startup.

[0005] The imaging device 20 captures an image of a part to be managed,
The image data is transferred to the image memory 1 through the input port 14.
3A, and stores image data for one screen in the image memory 13A. FIG. 5 shows an example of an image captured in the image memory 13A. In this example, an example of an indoor image is shown. Here, the presence of the boundary line L between the wall and the floor is detected,
Further, an example of detecting the slope a and the intercept b of the boundary line L will be described.

[0006] When the image data is taken into the image memory 13A, the edge detecting means 13C is activated. The edge detection means 13C detects a portion in the image where the density value changes rapidly (this portion is generally called an edge), and generates edge image data in which only the edge portion is extracted.
This edge image data is written into the image memory 13A.
Therefore, the image data written in the image memory is rewritten to the edge image data.

The binarizing means 13D is activated at the same time when the operation of the edge detecting means 13C ends. The binarizing means compares the brightness of each pixel of the edge image written in the image memory 13A with a reference value, and determines the brightness of each pixel as white or black.
Then, the binarized image data is written into the image memory 13A. Accordingly, the image data stored in the image memory 13A is normalized to only pixel data of, for example, “0” logic for displaying white and “1” for displaying black on the screen.

At the same time when the operation of the binarizing means 130 ends, the Hough transforming means 13E is started. Hough transform means 1
3E sequentially selects candidate points one by one from a set of black points (hereinafter referred to as candidate points) representing, for example, a boundary line L stored in the image memory 13A, and performs Hough transform for each candidate point. . As is well known, the Hough transform assumes a straight line in any direction passing through the candidate point P ₁ (see FIG. 6), and from this straight line to the origin on the image (one of the four corners of the image XY coordinate is set), a perpendicular is drawn, the inclination of the perpendicular is θ, the length of the perpendicular is ρ, and the candidate point P ₁
Are plotted on the θ-ρ plane shown in FIG. 7 when the inclination θ of the perpendicular and the length ρ of the perpendicular are changed when the inclination of the straight line passing through is changed. By repeating this Hough transform for each of the candidate points P ₁ , P ₂ , P ₃ ..., A plurality of curves can be obtained on the θ-ρ plane. Each data of this curve group is stored in the curve group memory 13B. The curve group intersects with a specific value of θ ₀ and a value of ρ ₀ .

The coordinates θ of this intersection point₀And ρ₀The frequency distribution search
The search means 13F detects the coordinates θ ₀And ρ₀Line extraction
The means 13G fits the XY coordinates as shown in FIG.
The slope a and intercept b of the straight line to be detected by_i
= Ax_i+ B. For more information on Hough transform,
Commentary is a questionnaire, for example, published by Technology Review Company
Basic Techniques of Science ”.

The image processing apparatus 10 calculates, for example, the slope a and the intercept b of the boundary line L, outputs the values of the slope a and the intercept b from the output port 15, and outputs the data to, for example, a controller (see FIG. (Not shown in FIG. 4).
The autonomous vehicle calculates the distance between itself and the wall, or the angle between the wall and the traveling direction (the direction of the optical axis of the imaging device 20), and uses the calculated data for automatic steering.

[0011]

A conventional image processing apparatus 1
At 0, the Hough transform is executed using the binary data fetched into the image memory 13A. The Hough transform is based on the X-Y coordinates set on the image, which is obtained by performing a Hough transform at a candidate point having a small XY value and a transform result at a candidate point having a large XY value. The degree of influence of the conversion error is different. That is, on the image shown in FIG. 9, when the candidate point P ₁ having a large XY value is Hough transformed and plotted on the θ-ρ plane, the candidate point P ₁ appears with a large amplitude as a curve A shown in FIG. 10. When contrast candidate point P ₂ where X-Y value has a smaller value to the Hough transform, the theta-[rho plane small amplitude curve as the curve B shown in FIG. 10 is obtained. The curve B having a small amplitude is greatly affected by the quantization error because the resolution of the perpendicular length ρ and the angle θ obtained from the XY coordinates of the image memory 13A is low. Therefore, if the effect of the quantization error is exaggerated and the curve B is expressed, the curve B changes as shown in the figure, and it is understood that the effect is greatly affected by the quantization error.

An object of the present invention is to provide an image processing method and an image processing method capable of performing Hough transform on pixel data stored in an image memory at a pixel position having a smaller XY value while reducing the influence of a quantization error. A high-precision image processing apparatus using this image processing method is proposed.

[0013]

According to the present invention, Hough transform is performed on candidate points scattered in an image taken into an image memory, and a straight line portion having the largest cumulative number of straight lines connecting candidate points scattered on the image by the Hough transform. In an image processing method for detecting an image, an XY coordinate having an origin at one of the four corners of the image is set for an image fetched into an image memory, and a small-value area including the origin and having small X and Y values is set. Image, and the divided image area is written in the image memory in an enlarged manner.
We propose an image processing method in which Hough transform is performed for each candidate point of the image stored in an enlarged manner.

According to the image processing method of the present invention, an image in a region having a small XY value in an image stored in an image memory is divided, and the divided image is enlarged to an image memory having the original number of pixels. Therefore, the XY value of the candidate point in the area where the XY value is small is corrected to the large XY value. As a result, it is possible to reduce the quantization error generated at the time of the Hough transform even for the candidate points in the region having a small XY value. Therefore, there is obtained an advantage that highly accurate Hough transform can be executed.

The present invention further proposes a high-precision image processing device using this image processing method. A high-precision image processing apparatus according to the present invention detects an edge of an image memory storing an image captured by an imaging device, and detects a portion in the image stored in the image memory where a density value changes sharply as an edge. Edge detecting means for extracting only a portion as an image, binarizing means for normalizing the edge detected image detected by the edge detecting means to pixel data having binary values of white and black, and the binarizing means For the processed image, an orthogonal coordinate system having an origin at one of the four corners of the image is set, and an image dividing unit that divides an image region of a small value region including the origin on the coordinate, and an image dividing unit An enlarged image memory for enlarging and storing the divided image area in an image memory having the same number of pixels as the image memory; and dispersing the image stored in the enlarged pixel memory and the image left in the image memory. Huff transform means for sequentially performing Hough transform on each candidate point to be calculated and calculating a curve on the θ-ρ plane, and coordinates at which the cumulative intersection number of the curve group on the θ-ρ plane converted by the Hough transform means becomes maximum. And a line having the largest cumulative number of lines connecting candidate points scattered in the image divided by the coordinate values obtained by the frequency distribution searching unit and the image left in the original image memory. And a straight line extracting means for calculating the slope a and the intercept b.

According to the high-precision image processing apparatus of the present invention, since the straight line is extracted by the Hough transform having a small quantization error, the slope a and the intercept b of the straight line can be obtained with high accuracy. In addition, an image area in which the values of X and Y on the rectangular coordinates are small is divided from the original image, and the divided image is simply enlarged and stored in an image memory having the same number of pixels as the original image memory. Therefore, it is not necessary to prepare a specially large image memory. Therefore, there is an advantage that the scale of the apparatus is not increased and the accuracy can be improved without increasing the cost.

[0017]

FIG. 1 shows an embodiment of a high-precision image processing apparatus according to the present invention. By describing the configuration and operation of this high-precision image processing apparatus, the image processing method according to the present invention will be described together. Reference numeral 30 shown in FIG. 1 indicates a high-precision image processing device according to the present invention. The high-precision image processing apparatus 30 according to the present invention has an image dividing means 31 and an enlarged image which enlarges and stores an image divided by the image dividing means 31 in addition to the configuration of the conventional image processing apparatus 10 shown in FIG. It is characterized by a configuration in which a memory 32 is provided.

The image dividing means 31 sets orthogonal coordinates (hereinafter referred to as XY coordinates) with the origin at one of the four corners of the image for the image fetched into the image memory 13A.
On the Y coordinate, an image of an area having a small XY value including the origin is extracted and divided. FIG. 2 shows this state. The entire area of the image memory 13A and A _0, sets the X-Y coordinates of the lower left corner as the origin O. On the X-Y coordinate, taken out by dividing the image X-Y value is included in a small small value area A _1. Candidate points are an X-Y value is smaller small value range _{A 1 P 11, P 12,} P 13 ... are included. The small value area A ₁
Are left in the image area A ₀ except for the candidate points P ₀₁ , P ₀₂ ,
P ₀₃ ... are included.

When the image of the small value area A ₁ is divided and taken out, the image dividing means 31 delivers the divided image data to the enlarged image memory 32. The enlarged image memory 32 has a pixel capacity equivalent to that of the image memory 13A, and enlarges image data extracted and divided (increases pixels by interpolation to match the same number of pixels as the number of pixels of the enlarged image memory 32).
And memorize.

An image is taken into the enlarged image memory 32.
And the edge detection means 13C is activated, and the edge detection image
In the enlarged image memory 32 and the original image memory 13A.
You. When the edge detecting means 13C completes the operation, it is binarized.
The means 13D is activated, and the enlarged image memory 32 and the original image
The image in the memory 13A is binarized. The binarization process is completed
Upon completion, the Hough transforming means 13E is activated and the enlarged image
Each candidate point P included in the memory 32₁₁, P₁₂, P₁₃…
Perform the Hough transform. By this Hough transform, FIG.
The group of curves shown is obtained. On the other hand, the original image memory 13A
Candidate point P ₀₁, P₀₂, P₀₃Huff transform is applied to ...
Then, the curve group shown in FIG. 3B is obtained and stored in the curve group memory 13B.
Remember.

When the curve group is obtained, the frequency distribution search means 13F is activated to detect the coordinates (θ ₀₁ , ρ ₀₁ ) and (θ ₀₂ , ρ ₀₂ ) of the maximum number of intersection points of each curve group. 3A and 3B, the candidate points P ₁₁ , P ₁₂ , P ₁₃ ... Stored in the enlarged image memory 32 are the candidate points P ₀₁ , P ₀₂ on the image left in the original image memory 13A. , P ₀₃ ... (The distance from the origin 0 is a close value), the curve group obtained by the Hough transform is also a curve having substantially the same amplitude. That is, the degree of influence of the quantization error on the curve group obtained from the candidate points P ₁₁ , P ₁₂ , P _13, ... Stored in the enlarged image memory 32 depends on the candidate points P ₀₁ , P remaining in the original image memory 13A. ₀₂ , P ₀₃ ..., And the candidate points of the small value area A ₁ , P ₁₁ , P ₁₂ , and P ₁₃ , can reduce the influence of the quantization error on the Hough transform. In particular, the coordinates (θ ₀₁ , ρ ₀₁ ) of the intersection of the curve group (curve group obtained from the enlarged image memory 32) shown in FIG. 3A
Will be more reliable.

The coordinates (θ ₀₁ , ρ ₀₁ ) and (θ ₀₂ , ρ ₀₂ ) of the most frequent crossing points of the separately obtained curve group should be essentially the same value, but if they have different values, If the value of the difference is not abnormally large, the intermediate value between the two is set to the final value (θ
_0, ρ ₀ ). When the final value (θ _0, ρ ₀ ) is determined, θ ₀ and ρ ₀ are transferred to the straight line extracting means 13G,
The slope a and the intercept b of the straight line to be obtained are obtained.

In the above-described embodiment, the number of divisions of an image is 1. However, the number of divisions is not limited to 1. It can be easily understood that the image can be divided into a plurality of areas in the order of XY values. Like.

[0024]

As described above, according to the present invention, the image of the small value area on the XY coordinates is enlarged to set the candidate points to the X-Y coordinates.
Since the image is moved to the high value area of the Y value and the Hough transform is executed in that state, the Hough transform of the image of the small value area of the XY value can be performed with high accuracy. Therefore, it is possible to improve the detection accuracy of the straight line, and to obtain an advantage that, for example, the automatic operation of the automatic traveling vehicle can be executed with high accuracy.

[Brief description of the drawings]

FIG. 1 is a block diagram for explaining an embodiment of a high-precision image processing device according to the present invention.

FIG. 2 is a diagram for explaining the operation of the high-precision image processing device shown in FIG.

FIG. 3 is a diagram similarly illustrating the operation of FIG. 1;

FIG. 4 is a block diagram for explaining a conventional image processing apparatus.

FIG. 5 is a diagram illustrating an example of an image to be subjected to image processing by a conventional device.

FIG. 6 is a view for explaining the principle of Hough transform used for image processing.

FIG. 7 is a view similar to FIG. 6;

FIG. 8 is a view similar to FIG. 6;

FIG. 9 is a diagram for explaining inconvenience of the conventional technique.

FIG. 10 is a view similar to FIG. 9;

[Explanation of symbols]

 Reference Signs List 11 central processing unit 12 read-only memory 13 rewritable memory 13A image memory 13B curve group memory 13C edge detecting means 13D binarizing means 13E Hough transforming means 13F frequency distribution searching means 13G straight line extracting means 14 input port 15 output port Reference Signs List 20 imaging device 30 high-precision image processing device 31 image dividing means 32 enlarged image memory

Claims (2)

[Claims] 1. An image processing method for performing a Hough transform on candidate points scattered in an image fetched into an image memory and detecting a straight line portion having the largest cumulative number of straight lines connecting the candidate points scattered on the image by the Hough transform. In the image captured in the image memory, the rectangular coordinates having the origin at one of the four corners of the image are set, and the image area of the small value area including the origin of the rectangular coordinates is divided from the image, and the image is divided. An image processing method, wherein an image area is enlarged and written into a memory having the same number of pixels as the image memory, and each candidate point of the enlarged and stored image is Hough-transformed. 2. A. B. an image memory for storing an image captured by the imaging device; B. edge detecting means for detecting a portion of the image stored in the image memory where the density value changes sharply as an edge, and extracting only the detected edge portion as an image; B. binarization processing means for normalizing the edge detection image detected by the edge detection means to image data having binary values of white and black; Image division in which an image processed by the binarization processing means is set with orthogonal coordinates having one of four corners of the image as an origin, and an image area of a low value area including the origin of the coordinates is divided from the image. Means; B. an enlarged image memory that enlarges and stores the image area divided by the image dividing means in an image memory having the same number of storage pixels as the image memory; G. Hough transform means for sequentially performing Hough transform on candidate points scattered in the image stored in the enlarged image memory and the image left in the image memory and calculating a curve on the θ-ρ plane; Frequency distribution search means for obtaining θ ₀ and ρ _{0 at} which the cumulative number of intersections of the curve group on the θ-ρ plane converted by the Hough transform means becomes maximum; The inclination coefficient a and the intercept b of the line having the largest cumulative number of lines connecting candidate points scattered in the divided image and the image left in the image memory based on θ ₀ and ρ _{0 obtained} by the frequency distribution search means are obtained. A high-precision image processing apparatus comprising: