JPH0799544B2 - Line segment extraction device - Google Patents

Description

Detailed Description [Table of Contents] Outline Industrial field of application Conventional technology Problems to be solved by the invention Means for solving problems (FIG. 1) Action Example (a) of one embodiment Description of configuration (Figs. 2, 3, 4)
(B) Description of operation of one embodiment (FIGS. 5, 6, and 7)
(C) Description of Other Embodiments Effect of the Invention [Outline] Calculation of a mapping function in a line segment extraction device that calculates a mapping function for image contour point coordinates and creates a histogram to extract line segment information. By connecting the section, the buffer, and the histogram creating section in series, the calculation of the mapping function and the creation of the histogram can be performed in parallel.

[Industrial application field]

The present invention relates to a line segment extracting device for extracting line segment information from an original image by mapping, and more particularly to a line segment extracting device capable of creating a histogram at high speed.

In recent years, object recognition technology has been actively developed in order to realize a function similar to that of the human eye by a machine.

In the recognition of such an object, a method of recognizing the object by using the image as an image and extracting information such as characteristics of the object from the image is used.

For this reason, it is generally necessary to pick up an object by an image pickup means such as a television camera, take in an original image, and perform electrical processing based on this image to extract information.

On the other hand, such an original image is relatively easy to extract the structural lines of the original image if it is a binary image processing subject to strong restrictions on the background and proof conditions, but under normal reflected light with less restrictions The image itself becomes a grayscale image, which makes it necessary to perform contour extraction processing from the grayscale image.

A problem in this contour extraction processing is that the original image contains a contour discontinuity or something other than the contour component, and a solution for how to extract the contour from such an unclear original image is desired. ing.

As a method for extracting such contours, Hough has been used conventionally.
Some use mapping by transformation and some use spherical mapping. The Hough transform is used as a method of concentrating the certainty in extraction of a straight line or a line segment of an input image. FIG. 8 shows an explanatory diagram of the Hough transform. In the so-called Hough transform, a point (xi, yi) on the image plane as shown in FIG. 8 (A) is expressed as ρ = xicosθ + yisinθ in the parameter plane as shown in FIG. 8 (B). Is converted into a sine wave (map function) represented by. Here, if a sine wave corresponding to each point on a straight line as shown in FIG. 8 (A) is plotted on the parameter plane and a histogram is created, two points (ρ
If you consider only positive ones, you can intersect at 1 point. The rotation angle θ ₀ and the height ρ ₀ at this point are the length ρ ₀ of the vertical line and the rotation point θ when the line perpendicular to the origin is drawn with respect to the straight line formed by each point in FIG. 8 (A). _Since it indicates ₀ , the information of the straight line formed by each point can be extracted by this. That is, it is possible to extract a straight line or a line segment, reproduce a line segment having a break or distortion, and the like.

[Conventional technology]

As a device for extracting contours, that is, line segments, by such mapping, the above-mentioned mapping function value ρ is calculated from the contour point coordinates (xi, yi) obtained from the original image as shown in FIG. 9 (A). It consisted of a processor CPU that calculates and creates a histogram and a histogram memory HM in which the histogram is written.

As shown in FIG. 9B, this processor CPU calculates a corresponding series of mapping function values ρ _{0 to} ρ ₅₁₁ from the contour point coordinates (x ₀ , y ₀ ) in the Hough transform, and calculates the calculated function. The process of creating and writing the histogram in the histogram memory HM from the values is alternately performed to complete the histogram for one coordinate in the histogram memory HM.

[Problems to be solved by the invention]

In such a line segment extraction device, the accuracy of the histogram increases as the mapping function ρ corresponding to the coordinates of the contour points is finely divided to generate the function value, and therefore the maximum point detection on the accurate histogram memory is performed. Is possible.

For example, if one mapping function ρ is divided into 512 as in the example of FIG. 9, line segment detection can be performed with an accuracy of within 1 pixel for an input image of 256 × 256 pixels. In order to detect such a high-precision maximum point, it is necessary to calculate the function value 512 times and create a histogram for each contour point. In the case where both calculation and histogram creation are performed alternately, there is a problem that it takes time on the order of minutes for one coordinate and the time required for line segment extraction becomes long.

Therefore, it is an object of the present invention to provide a line segment extracting device that can create a histogram from contour point coordinates at high speed.

[Means for solving problems]

 FIG. 1 is an explanatory view of the principle of the present invention.

In the figure, 1 is a function value calculation unit, which is a contour point coordinate (xi, yi)
From which the function value of the corresponding mapping function is calculated, 2 is a buffer, and the function value calculated by the function value calculation unit 1 is stored, 3 is a histogram creation unit, which is stored in the buffer 2. The histogram creation calculation is performed from the function value and written in the histogram memory.

Therefore, in the present invention, the function value calculation unit 1 and the histogram creation unit 3 exist separately and are connected in series via the buffer 2.

[Action]

In the present invention, since the function value calculation unit 1 and the histogram creation unit 3 are provided separately, the function value calculation and the histogram calculation can be executed in parallel.

Moreover, since the calculation unit 1 and the creation unit 3 are connected via the buffer 2, even if there is a difference in calculation time, the calculation unit 1 and the creation unit 3 can operate in parallel.

Therefore, the process from contour point coordinates to histogram creation can be speeded up.

[Embodiment] (a) Description of Configuration of One Embodiment FIG. 2 is a configuration diagram of one embodiment of the present invention.

In the figure, the same components as those shown in FIG. 1 are designated by the same symbols, 4 is a television camera, which captures a target image and converts it into an image signal, and 5 is a contour extraction unit. ,
An A / D converter 50 for extracting contour points from the image signal from the TV camera 4 and generating contour point addresses. The image signal of the TV camera 4 is converted from analog / digital into 8-bit image data. And the A / D converter 50
The spatial differentiation circuit 51 for spatially differentiating the output of the above to remove other than the contour (edge), the extreme value processing circuit 52 for processing the output of the spatial differentiation circuit 51 by the thin line, and the output of the extreme value processing circuit 52 And a contour address generation circuit 53 for generating the address 6 is a speed adjustment buffer, which temporarily stores the contour point address from the contour extraction unit 5, 7 is a histogram memory, in which a histogram is written by the histogram creation unit 3, for example, 256K
A host processor having a bit capacity, 8 is a host processor, a speed adjustment buffer 6, a function value calculation unit 1, and a buffer 2
In addition to controlling the histogram creation unit 3, the contents of the histogram memory 7 are analyzed to extract line segments.

FIG. 3 is a block diagram of the function value ρ calculator in the configuration of FIG.

In the figure, 10 is an address counter, which counts clocks to generate an address n in the θ direction 10 to π in Hough conversion, 11 is a cos θ table, and the value of cos θ corresponding to each address n is used as a table. Have, 12
Is a sin θ table, and sin θ corresponding to each address n
Having values of as a table, 13a, 13b, 13c, 13
d is an input register, and the input register 13a is a buffer 6
For storing the contour point coordinates xi from the input register 13b
Is the one that stores the cosθn value from the cosθ table 11, the input register 13c is the one that stores the contour point coordinates yi from the buffer 6, and the input register 13d is the si from the sinθ table 12.
The nθn value is stored.

14 and 15 are multipliers, and the multiplier 14 is an input register 13a.
The coordinate xi of x and the cos θn value of the input register 13b are multiplied, and (xi
・ Cos θn) is output, multiplier 15 is input register 1
Multiply the coordinate yi of 3c by the sin θn value of the input register 13d,
(Yi · sin θn) is output, 16a and 16b are output registers, the output register 16a is the output of the multiplier 14 (xi · cos θn) is set according to the clock, and the output register 16b is the clock. Output of the multiplier 15 (yi
.Sin θn) is set, 17 is an adder,
The contents of each output register 16a, 16b are added, and ρn = (xi · c
os.theta.n + yi.sin.theta.n) is output, and 18 is a write register for setting the output .rho.n of the adder 17 in accordance with the clock and writing it in the data buffer 2.

Therefore, the function value calculation unit 1 is configured to calculate the Hough transform function ρn, and the input contour point coordinates (xi, yi)
On the other hand, the Hough transform function value ρn of each address n (= 0 to 511) is configured to be calculated by hardware,
The address n and the function value ρn are sequentially written in the buffer 2.

FIG. 4 is a configuration diagram of the histogram creation unit 3 in the configuration of FIG.

In the figure, 30 is an address generator, which generates an address n from the buffer 2 and a function value ρn as addresses of the histogram memory 7 and accesses the histogram memory 7, and 31 is an increment circuit from the histogram memory 7. One which adds 1 to the 8-bit data read by the address of the address generation unit 30 and outputs it as data to the histogram memory 7, 32 is a memory controller, which reads / writes the histogram memory 7 according to the address clock. The write control is performed and the operation of the increment circuit 31 is controlled.

Therefore, the histogram creating unit 3 is configured to read the data at the corresponding position in the histogram memory 7 using the address n and the function value ρn as an address, add "1" to this data, and write again at the same position. In the histogram memory 7, 8-bit data is 512 vertical and 512 horizontal.
It is composed of a memory that can accommodate.

(B) Description of operation of one embodiment FIG. 5 is an illustration of ρ calculation operation by the configuration of FIG. 3, FIG. 6 is an illustration of histogram creation operation by the configuration of FIG. 4, and FIG. Is.

The original image (grayscale image) input from the TV camera 4 is
The contour point address (xi, yi) is input to the contour extraction unit 5, the contour point address (xi, yi) is extracted in the well-known manner, and is output to the speed adjustment buffer 6. The speed adjustment buffer 6 is composed of a first-in first-out memory and sequentially stores the extracted contour point addresses (xi, yi). On the other hand, the function value calculation unit 1 receives the contour point address (xi, yi) from the speed adjustment buffer 6 and operates as shown in FIG.

First, the clock cycle T is set according to the multiplication times of the multipliers 14 and 15. Under the condition of the same element technology, the multiplication time is longer than the addition time and the table access time in the configuration shown in FIG. 3, so it is set in accordance with the maximum multiplication time.

Then, each function value ρn is calculated by dividing 0 to π of the Hough transform parameter θ into 512 equal parts.

Therefore, the formula (1) is The cos θ table 11 and the sin θ table 12 store 512 cos θ and sin θ values corresponding to the address n.

As shown in FIG. 5, the contour point address (x
i, yi), the clock causes the registers 13a, 13c
These are set, and cos θ table 11 to cos θ ₀ and sin θ table 12 to sin θ ₀ are set in registers 13b and 13d, respectively.

Therefore, each of the multipliers 14 and 15 has (xi · cos θ ₀ ), (yi · sin
θ ₀ ) is calculated, and the multiplication output is set in the output registers 16a and 16b at the next clock.

Further, the multiplication output set in the registers 16a and 16b is the adder 1
The value is added at 7, and is set in the write register 18 at the next clock.

Therefore, with 3 clocks, the address “0” for (xi, yi)
The function value ρ ₀ of

Since the ρ calculator 1 sequentially performs the operation serially by the step of the address counter, the function value is ρ for each clock.
₀ , ρ ₁ , ρ ₂ ... ρ ₅₁₁ are output, so that the time for generating a series of 512 function values for one contour point address (xi, yi) is (512 + 3) × T. The calculated function value ρn is stored in the data buffer 2 together with the address n, as shown in FIG. On the other hand, in the histogram creation unit 3, the address generation unit is generated every time the address clock arrives.
The n, ρn of the data buffer 2 is read from 30 and the histogram memory 7 is accessed using this as an address. Since the memory controller 32 generates the write enable signal ▲ ▼ shown in FIG. 6 according to this address clock, the data Dn at the accessed address comes first from the histogram memory 7 like the memory I / O data shown in FIG. Are read out and applied to the increment circuit 31, and the increment circuit
"1" is added at 31, and the output data D of the increment circuit 31 is changed by the low level of the write enable signal ▲ ▼.
_{n + 1} is written to the same address.

Therefore, as shown in FIG. 7, after the 512 ρn are calculated by the ρ calculation unit 1, the 512 ρn are written in the histogram memory 7 by the histogram creation unit 3. During this writing ρ
The calculation unit 1 calculates ρn for the next contour point coordinate (x1, y1), and accordingly, the function value calculation and the histogram creation operate in parallel.

Therefore, the conventional time required to create a histogram is the sum of the ρ calculation time Tρ and the histogram creation time Tn, whereas in the present invention, the longer time of Tρ and Tn is sufficient.

In addition, the calculation time of the ρ calculation unit 1 is significantly reduced. That is, the conventional calculation time is 2 table accesses and 2 multiplications.
The number of times is 5 times, that is, one time and one addition time, and if the multiplication time is added and the table access time Ta is set to be twice, 512 × 7 × Ta is obtained. Get faster

When the histogram is written in the histogram memory 7 in this way, the host processor 8
Is read, the address of the maximum value is searched as an intersection point, and the angle θ ₀ and the distance ρ ₀ of the line segment described above are reproduced and extracted by the address of the maximum value.

(C) Description of Other Embodiments In the above embodiments, the Hough transform function was described as an example of the mapping function, but other transform functions may be used.

The embodiments of the present invention have been described above, but the present invention can be modified in various ways according to the gist of the present invention, and these modifications are not excluded from the present invention.

〔The invention's effect〕

As described above, according to the present invention, since the calculation of the mapping function and the creation of the histogram can be performed in parallel, even if the time required to create the histogram in the line segment extraction is shortened,
This has the effect of enabling high-speed line segment extraction. In particular, since highly accurate detection is performed, its effect is great when a large number of mapping function values are generated, and it greatly contributes to speeding up the line segment extraction processing.

[Brief description of drawings]

FIG. 1 is a diagram for explaining the principle of the present invention, FIG. 2 is a block diagram of an embodiment of the present invention, FIG. 3 is an internal block diagram of a function value calculation unit in the configuration of FIG. 2, and FIG. 5 is a configuration diagram of a histogram creating unit of FIG. 5, FIG. 5 is an explanatory diagram of a function value calculation operation according to an embodiment of the present invention, FIG. 6 is an explanatory diagram of a histogram creating operation according to an embodiment of the present invention, and FIG. FIG. 8 is a diagram for explaining the Hough transform, and FIG. 9 is a diagram for explaining a conventional technique. In the figure, 1 ... Function value calculation unit, 2 ... Buffer, 3 ... Histogram creation unit, 5 ... Contour extraction unit, 7 ... Histogram memory.

Claims (1)

[Claims] 1. A line segment extraction for calculating a mapping function value for contour point coordinates of an image, creating a histogram according to the calculated mapping function value, and extracting line segment information indicated by the contour point coordinates from the histogram. In the apparatus, a mapping function value calculation unit that calculates a mapping function value from the contour point coordinates, a buffer that stores the mapping function value calculated by the mapping function value calculation unit, and a mapping function value from the buffer to a histogram memory. A line segment extraction device comprising: a histogram creation unit that creates a histogram, wherein the mapping function value calculation operation by the mapping function value calculation unit and the histogram creation operation by the histogram creation unit are performed in parallel.