JPS63214888A - Picture processing method - Google Patents

Abstract

PURPOSE:To measure an end position of an object accurately while excluding causes of error due to a change in the lighting or the shape of the end by applying statistical processing to the picture edge. CONSTITUTION:Two-dimension picture data of an object 100 is obtained by an image pickup means and the two-dimension picture data is divided into plural areas in two different directions, e.g., x, y directions. Then the change in the picture data sampled at two adjacent areas is detected as to all the regions as to the two directions x, y. The distribution of frequency of sample numbers where the change in the picture data is a prescribed threshold value or over as to two directions x, y is obtained and the position of the end of the object 100 is detected based on the distribution of frequency.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention relates to an image processing method for detecting the end position of an article having a substantially rectangular shape, such as a headlight of an automobile.

(Prior Art) Conventionally, in the assembly of automobiles, after the headlights are assembled, it is necessary to perform adjustment inspection to ensure that the optical axis is within a predetermined standard range. This forward axis adjustment is performed so that the brightest point or the brightness/darkness boundary line of the headlight irradiation light is within a reference range.

Regarding this optical axis adjustment, a technology has been proposed that improves the precision of optical axis adjustment by optically image processing the light distribution pattern of the irradiated light, as seen in, for example, Japanese Patent Laid-Open No. 59-24232. ing. In this prior example, a bright and dark boundary line (estimated sharp cut line) located at a certain relative position from the center of gravity of the light distribution pattern of the irradiated light at a certain luminous intensity or higher is estimated and found for the same car model, and the The optical axis of the headlights is adjusted so that the boundary line between bright and dark is included.

In addition, as a technology for adjusting the optical axis of the same headlight,
Patent application Sho 6l-1tA66 from the patent applicant for the claimed invention
There is a proposed technology like No. 15. The aforementioned Japanese Patent Application Publication No. 59-2
In view of the fact that No. 4232 is based on the unfounded premise that there is a certain relative relationship between the brightest point and the bright/dark boundary line in the same car model, the above proposed technology is based on the actual brightest point for each car. We are trying to achieve highly accurate optical axis adjustment by finding the relative positional relationship between the point and the contrast between bright and dark boundaries.

(Problem to be solved by the invention) However, the above two optical axis adjustment technologies affect the headlight position (direction of the headlight beam), such as the height of the vehicle and the left-right position of the vehicle body. It is assumed that you know the elements accurately. Due to differences in tire pressure, etc., the height of the vehicle may vary even for the same model, and as a result, even if the brightest point of the headlight irradiation is detected, the standard brightness/darkness boundary line that should be compared with it cannot be determined. This is because it becomes

Therefore, the above-mentioned Japanese Patent Application Laid-open No. 59-24232 uses a TV camera etc. to measure the vehicle height and predict the headlight position, and the above proposed technology also uses a TV camera etc. to detect the upper and inner sides of the headlights. to determine the headlight position.

However, when measuring the headlights or vehicle height using an imaging means such as a television camera, the edges of the headlights, the edges of the vehicle body, etc. cause halation, diffused reflection, etc., making it easy to make erroneous measurements.In other words, the adverse effects of illumination It is easy to accept.

The above problem is not limited to edge detection of headlights, vehicle bodies, etc., but also occurs when detecting edge positions of general articles by image processing.

Therefore, the present invention was proposed in order to eliminate the drawbacks of the above-mentioned conventional examples, and its purpose is to propose an image processing method that can capture image data of an article and accurately measure the position of the article. It is.

(Means and operations for solving the problems) The configuration of the image processing method of the present invention for achieving the above-mentioned problems is as shown in FIG. , the two-dimensional image data is
For example, as shown in the figure, the division step is to divide into multiple regions in the x and a change point detection step for detecting a change point; a distribution step for determining a frequency distribution in the two directions of the number of samples in which the change is greater than or equal to a predetermined threshold; and an end portion for detecting the position of the end of the article based on the frequency distribution. It consists of a detection step.

(Examples) Examples according to the present invention will be described below with reference to the accompanying drawings.

FIG. 1 is a diagram illustrating the principle of a basic embodiment of the present invention for confirming the position of the end of an article. When an image of an article 100 having a substantially rectangular shape is captured by a camera having a CCD area sensor (or line sensor), etc., two-dimensional multi-valued image data as shown in FIG. 2(b) is obtained. One pixel of these multivalued image data usually corresponds to one pixel of a COD sensor, and expresses density changes of, for example, 8 bits and 256 gradations. If the surface of the article 100 is covered with the same material, the values of the two-dimensional multivalued image data will have only slight changes except for the edge portions.
Since the brightness of the end portion changes greatly, the multivalued image data value thereof changes greatly. This is the image edge.

This change is usually perceived as a darkening of the brightness at the edges. Furthermore, in the image of this end portion, the above-mentioned image edge position varies due to halation, diffused reflection, and the like. This state is shown in FIG. 1(b). In FIG. 1(b), multivalued image data values are three-dimensionally shown in the z-axis direction within the image plane indicated by x and y coordinates.

Regarding multivalued image data having such a distribution, changes in image data values are detected for two adjacent pixels in each of the X direction and the X direction. FIG. 1(C) three-dimensionally shows the difference in image data values between two pixels adjacent to each other in the X direction for each line in the X direction in two axial directions. Similarly, FIG. 1(d) shows the difference in image data values between two pixels adjacent to each other in the Y direction in two axial directions in a three-dimensional and one-dimensional manner. Points where this difference is large should correspond to image edges at the ends of article 100.

From these two distributions, the number of sample points for which the difference is greater than or equal to a predetermined value is accumulated, and a frequency diagram (histogram) is created.
This is shown in Figure 1(e).

Cf), the location of the sample point with the peak value is considered to be the center of the image edge corresponding to the edge of the article. That is, ■: By performing the above statistical processing, it is possible to exclude the effects of illumination such as halation and diffused reflection, and even if the edges of the article 100 do not form a straight line, the edges can recognize the location of

■: Even if the surface of the article 100 is covered with a uniform substance, there may be some variation in the image data, but there is a difference in the multivalued image data between two adjacent pixels. By statistically processing the above, noise caused by the above-mentioned variations can be absorbed.

In addition, in the embodiment shown in FIG. 1 above, change point detection and
Although the creation of the frequency distribution is explained as a separate process,
, and to obtain the frequency distribution in the X direction, fix the X address counter value of this image memory, read out the image data while incrementing the X address counter, and calculate the difference between them. The number of items exceeding a predetermined value is counted, and this operation is repeated while incrementing the X address counter. The same applies to the frequency distribution in the X direction. In this way, high-speed processing becomes possible.

However, performing image processing on multivalued image data as in the basic embodiment shown in FIG. 1 requires a long processing time. Therefore, in a modified embodiment in which the present invention is applied to the optical axis adjustment of an automobile headlight, which will be explained below in Fig. 2, binary image data is used instead of multivalued image data in order to increase the processing speed. . In this embodiment, as will be described later, in view of the fact that the binary image data has a small tolerance range for the binarization level and is susceptible to noise, the binarization level (Jl
) is made variable during the edge detection process.

In the embodiment described below in FIG. 2, the image processing method according to the present invention is applied to the confirmation of the headlight position when adjusting the optical axis of the headlight of an automobile. Here, in the headlight position confirmation method of the present embodiment, an article such as a headlight has substantially the same brightness in parts other than its ends, and
On the other hand, the fact that the brightness decreases (or increases) at the edge portion is utilized. FIG. 2 shows the overall configuration of the front axis adjustment system, and the front of the automobile 1 that has been brought into the headlight optical axis inspection line is illuminated with light from left and right headlights 2a and 2b, respectively. In the vicinity of the left and right screens 3a and 3b, there are left and right first television cameras 4 that take images of the left and right headlights 2a and 2b of the automobile 1.
a, 4b (CCD camera) are installed, while the screen 3a.

Second television cameras 5a and 5b (CCD cameras) on the left and right sides are installed to take images of the light distribution pattern A (FIG. 8) of the irradiated light on the screen 3b. These television cameras output an analog image signal obtained by scanning an object to be imaged using a so-called rusk scan method to an image processing device 6 together with a horizontal synchronization signal and the like. Further, a signal from an operation panel 7 for inputting information such as the vehicle type is similarly inputted manually to an image processing device 6, and an output signal from this image processing device 6 is installed near the left and right headlights 2a, 2b. display device 8a
, 8b.

FIG. 2B is a block diagram of this image processing device 6. C
PU10 is a central processing unit that controls the entire image processing device 6. TV cameras 4a, 4b, 5a, 5
The analog video signal for each horizontal line from b
By the analog multiplexer 15 according to the instruction of 0,
An appropriate one is selected, converted into digital multi-valued image data, and stored in the image memory 13. 11 is a binarization circuit, and 12 is a change point detection circuit.

FIG. 3A shows an example of the binarization circuit 11, which serially compares the multi-value image data in the image memory 13 one pixel per line with the binarization level by a digital comparator (CMP) 20. , binarize. This sentence is cpu
io, and its value is variable as described below. FIG. 3B shows an example of binarized image data for each pixel.

The change point of the image data binarized by the CMP 20 is detected by the change point detection circuit 12. FIG. 3C is a circuit diagram of this circuit 12.

21 is a flip-flop, which is used to delay one pixel in order to detect a change in the image data value of two pixels adjacent in each of the x and y directions using an exclusive OR gate (EXOR) 2e2. The pixel clock in the figure is a clock for obtaining timing for each pixel. In this way, the gate 23 outputs "1" for each pixel position where the binary image data value changes from 1" to "0'" or from "O" to "1". For convenience, this two-dimensional distribution of change points detected in the X direction, which indicates a data string in which change points are extracted in the X direction from image data, is referred to as "X direction change point distribution."

In the basic embodiment shown in FIG. 1, FIG. 1(c) corresponds to this "X-direction change point distribution". This change point detection is also performed in the X direction. This is also called the "X-direction change point distribution" and corresponds to (d) in FIG. 1 in the basic embodiment. .

Figure 4A shows the appearance of the headlight part of the car 1.
Although FIG. 4A shows the periphery of the headlight 30 by drawing its outline for convenience of explanation, it is actually expressed as the shading of the image. Now, when such an object is imaged by the television cameras 4a and 4b, and binarized by the binarization circuit 11, and the change point is detected by the change point detection circuit 12, the images (a) and (b) in FIG. ) can be obtained. In addition, (a) of FIG.

(b) shows the above-mentioned "X-direction change point distribution" and "X-direction change point distribution" in the vicinity of the headlight 3o, with the change points (image edges) shown in black and the no-change points shown in white. It is something that

Therefore, cpuio adds the "1" on the first line in the X direction to the "X direction change point distribution" data, and calculates the cumulative value for each of the first line (x=1). as a numerical value, then find the degree value of the second line (x = 2),
Repeat this operation for all lines in the X direction, as shown in Figure 5 (
Obtain the frequency distribution in the X direction as shown in C). The same operation is performed for all lines in the X direction, and the frequency distribution in the X direction (
(d) in FIG. 5 is obtained. Furthermore, image edges are detected based on these frequency distributions, and headlight positions, vehicle body positions, etc. are further detected from the image edge positions. As shown in FIGS. 5(C) and 5(d), in the frequency distribution in each of the x and y directions, a peak point having a frequency greater than a certain set value is defined as an image edge.

By the way, some articles such as automobiles naturally have image edges in addition to the headlights 30. A plurality of such image edges are displayed as shown in FIGS. 5(C) and 5(d). In cases such as adjusting the optical axis of a car's headlights, the type of car is known in advance and its shape (distribution of image edges) is also known, so the distribution of image edges to be obtained can also be predicted. That is, the absolute position of the image edge in the vicinity of the headlight can be measured in advance for each vehicle type, and the distribution can be stored in a memory or the like. In this case, it is sufficient to memorize the positions and frequency values of individual peaks, the distances between peaks, and the like.

On the other hand, since the position of the automobile 1 carried into the system shown in FIG. 2A does not vary so much, some of the image edges obtained therefrom (for example, (C) in FIG. 5) match the known image edge distribution described above. In other words, the distribution of the obtained image edges and the distribution of known image edges are in a relative translation relationship, and as mentioned above, the absolute position of the known image edges is Since the above-mentioned parallel movement distance can be easily found by calculating the deviation between the peaks, even if the car 1 stops at a misaligned position, the edges of the image may be blurred due to the lighting conditions. Also, the absolute position of the end of the headlight 30 can be easily known.

Now, we will briefly explain how to adjust the optical axis after detecting the headlight position as described above. FIG. 6 is a flowchart of the control procedure for adjusting the optical axis. FIG. 7 is a flowchart of the procedure for measuring the headlight position described above.

FIG. 8 shows an optical axis adjustment instruction image displayed on the display devices 8a and 8b to assist the adjustment operator during optical axis adjustment. In this optical axis adjustment instruction image, the equilinear striped A is
The irradiation light of one headlight is projected onto the screens 3a and 3b, and the image processing device 6 calculates the images taken by the television cameras 5a and 5b, and connects the parts with equal brightness with equal straight lines. This is a visible image of the light distribution pattern.

P indicates the brightest point in the image.The position of this point P is also determined by subdividing the headlight irradiation image into two-dimensional pixel blocks, as shown in Japanese Patent Application No. 6l-11f41,113. Divide the image to obtain a histogram, find a threshold for binarization from this histogram, binarize it, find the center of gravity from the area of the "1" part of this binarized image, and use this center of gravity as the brightest point. Let it be P. L is an image in which a boundary where brightness actually changes suddenly in a headlight irradiation image is emphasized and visualized as a straight line image to help the adjustment operator. This L consists of a horizontal line part a and a diagonal line part A. S is the intersection point 0 In this example, a, b, S
If the positions of the headlights 2a and 2b are measured by the image processing method of the present invention and the optical axes of the headlights are properly adjusted, aLo obtained by the image processing device 6 is as follows. This is a reference brightness/darkness boundary line that should be located below this L. Therefore, the optical axis adjustment operator can adjust the optical axis with this L as a target.

In step S1 of FIG. 6, the type of vehicle 1 to be carried in is input from the operation panel, and the vehicle 1 is also stopped at a predetermined position on the optical axis adjustment line. By inputting the vehicle type, the image processing device 6 can learn the shape of the headlights, the photographing direction of the television camera, the peak distribution pattern that serves as a reference for the headlight image, and the like. In step S2, an image of the vicinity of the headlights is input. This is a process of importing multivalued image data obtained by the television cameras 4a, 4b into the image memory 13.First, in step S3, images of the headlights 2a, 2b are input. The display device 8a displays an instruction to turn on the light.
, 8b.

On the other hand, step S
2, the process proceeds to step S4 and detects the headlight position.The detailed procedure for detecting the headlight position is shown in FIG.
The binarization level at this time is unknown. In step S21, the change point detection circuit 12
Find the "X direction change point distribution" and "X direction change point distribution". Next, in step S22, (C) of FIG.

Obtain frequency distributions in both the X+3' directions as shown in (d). In step S23, image edges in each direction are detected from the two frequency distributions, and in step S24, the distance between peaks is determined. In step S25, the distribution of the known peak pattern described above is compared with the distribution of the peaks determined in step S24, and if the two patterns match at any relative position, the process proceeds to step 326. , calculate the relative distance between the distributions. On the other hand, if there is no matching pattern in step S25, the binary level sentence is changed by a predetermined amount a in step S30.

In step S31, it is determined whether the value of the sentence after this change is a meaningful value as a binarization threshold, and if YES, the process returns to step S20, the binarization is redone, and the above-mentioned flow is continued. repeat.

If NO in step S31, there is a problem such as the position of the car is too misaligned or the car height is too high.
Since the optical axis cannot be adjusted, this fact is displayed on the display device 8 as an error process. In this case, the operator would move the position of the car 1.

In this manner, the absolute position of the headlight within the optical axis adjustment system of FIG. 2a is detected by the headlight position detection subroutine of step S5. Therefore, the process returns to step S5 in the flowchart of FIG. At this point, the absolute positions of the headlights 2a and 2b are known, so if the headlights whose optical axes are adjusted within the standard are in that position, where on the screens 3a and 3b will the bright and dark boundary appear? The standard light/dark reference line LO, which is called ``Shutaku'', is calculated in step S5. Then, this Lo is displayed on the display device 8a,
8b.

On the other hand, when the headlights 2a and 2b are turned on in step S6, the above-mentioned image processing is performed in step S7, and A, L, P, a, b, etc. as shown in FIG. 8 are determined and displayed. In this case, as explained in Japanese Patent Application No. 67-/'7'6675, "0" in the X direction of the image data that has been subdivided into blocks of the headlight irradiation image and binarized is changed to "0" in the X direction. 1
Find the straight line a from the coordinate group of points that change to ``, for example, by the method of least squares.The straight line a is also found from the coordinate group of changing points in the X direction (or X direction).S is the intersection of a and b. Further, in step S8, the distance between P and 8 is found.As long as the headlights of the same car are used, this distance remains unchanged even if the optical axis is adjusted and moved.In step s9, the distance between P and 8 is determined. Determine whether the axis is within the specifications.This is done by the operator looking at the marks and Lo displayed on the display devices 8a and 8b and determining whether or not the axis is within the specifications. , the key that shows the judgment result (operation panel 7
This is done by the image processing device 6 determining the input (on the top). If it is within the standard, proceed to step S14,
The screen will go up and the adjustment is complete.

If it is outside the standard, the process proceeds to step s10, where the operator adjusts the optical axis. Next, in step Stt, images irradiated by the headlights are again taken into the image processing device 6 from the television cameras 5a and 5b, and a new position of the brightest point P is determined by the method described above. As mentioned above, since D is unchanged, it is possible to predict the brightness/darkness boundary line from P and D without the need for the calculations performed in step S7, and it is displayed on the display devices 8a and 8b. to be displayed. On the other hand, t
, o remains at the initially measured position. Then, the process returns to step S9 and the above-described operations are repeated.

Thus, if the image processing method of the present invention is applied to the optical axis adjustment of a headlight, the position of the headlight can be accurately detected by eliminating the effects of changes in illumination, differences in the shape of the end of the headlight, etc. As a result, optical axis adjustment that requires delicate adjustment can be performed accurately.

(Effects of the Invention) As explained above, according to the image processing method of the present invention, causes of error due to changes in the shape of the edge, illumination, etc. are eliminated by statistical processing of image edges in the edge position of the article. do,
Can be measured accurately.

[Brief explanation of the drawing]

Fig. 1 (a) to (f) are diagrams explaining the principle of the basic embodiment of the present invention, and Fig. 2A is a case in which the image processing method of the present invention is applied to an optical axis adjustment system of an automobile headlight. 2B is an internal block diagram of the image processing device 6 in the system shown in FIG. 2A, and FIGS. 3A and 3B are circuit diagrams of an example of a binarization circuit and a change point detection circuit, respectively. , 3C and 3D respectively show the above-mentioned binarization circuit. A diagram showing an example of an image data string obtained by the change point detection circuit, FIG. 4 is an external view showing the vicinity of the headlight of a car, and FIGS. FIGS. 6 and 7 are flowcharts showing the overall control procedure in the optical axis adjustment system, and FIG. FIG. In the figure, 1...Car, 2a, 2b, 30...Headlight, 3a, 3b...Screen, 4a, 4b, 5a, 5
b...TV camera, 6...Image processing device, 7...
・Operation panel, 8a, 8b...Display device, 10 ・CPU
, 11... Binarization circuit, 12-... Change point detection circuit,
13--Image memory, 14-A/D converter, 15-...
・°Analog multiplexer, 20...Digital comparator, 21...Flip-flop, 22-...Exclusive OR circuit, 23--A N D gate, A...
Light distribution pattern, P... brightest point, L... light and dark boundary line,
Lo: Reference brightness/darkness boundary line, D: Relative distance.

Claims (5)

[Claims] (1) An input step of obtaining two-dimensional image data of the article using an imaging means; A dividing step of dividing the two-dimensional image data into a plurality of regions in two different directions; a change point detection step of detecting a change in image data values sampled between two areas for all areas;
An image processing method comprising: a distribution calculation step of calculating a frequency distribution in a direction; and an edge detection step of detecting a position of an end of the article based on the frequency distribution. (2) The edge detection step includes a step of determining a sample point corresponding to the peak of the frequency distribution in one direction as an image edge in another direction, and identifying the image edge as an edge of the article. An image processing method according to claim 1. (3) the edge detection step includes a comparison step of comparing the spacing between sample points corresponding to two peaks for the frequency distribution in one direction with a known spacing between the edges of the article; Claim 1, characterized in that when the two intervals are unequal, the distribution calculation step to the edge detection step are repeated, and the distribution calculation step includes a step of changing the first threshold value. Image processing method described. (4) The input step includes a binarization processing step of binarizing the multivalued image data by comparing it with a predetermined second threshold value, and the edge detection step includes two values for the frequency distribution in one direction. a comparison step of comparing a spacing between sample points corresponding to a peak with a known spacing between edges of the article; edge detection from the binarization processing step when the two spacings are not equal; 2. The image processing method according to claim 1, wherein the steps are repeated, and the binarization processing step includes a step of changing the second threshold value. (5) The image processing method according to any one of claims 1 to 4, wherein the two directions are vertical and horizontal directions that are perpendicular to each other.