JP2002228423A - Tire detecting method and device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a tire detecting method and device for detecting the presence or absence, size, and location of a tire through the use of image processing on an input image without installation work of a foot board. SOLUTION: The image picked up by an image pickup device is inputted by image input S1 to perform field correction processing S2. In the field correction processing S2, movement displacements in lateral directions in the even and odd fields of an NTSC image from the image pickup device are automatically corrected on a vehicle moving in the lateral directions to compose a frame image. In noise removal S3, smoothing processing is performed for eliminating micro noise of a road surface. In edge detection S4, the contour edge of the tire is detected. In binarization processing S5, binarization processing is performed on the image to clarify the edge information of the tire. Contraction processing S6 is performed to remove the micro noise of the road surface. In tire detection pseudo Hough transform S7, the shape of the tire is defined as an ellipse to perform tire detection pseudo Hough transform processing. In tire recognition processing S8, a ballot table created in the tire detection Hough transform S7 is analyzed to determine the presence or absence, location, and size of the tire.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an input image obtained by imaging an area near a tire of a passing vehicle from an obliquely upper side with an imaging device.
The present invention relates to a tire detection method and a tire detection method for detecting a tire.

[0002]

2. Description of the Related Art Conventional tire detection methods include a method of measuring the number of axes using sensor information of a tread plate or the like, and a method of paying attention to a density pattern change in moving image processing when image processing is used. There are known detection methods based on tire pattern matching (for example, see Japanese Unexamined Patent Application Publication No.
No. 33525).

[0003]

However, there is a problem that the installation cost is high in the conventional method which requires the installation of the treads, and there is a problem that the treads cannot be installed due to the environment.

In the conventional method using image processing,
Depending on the imaging environment and the type of tire, a change in density pattern may not be remarkably observed, or even in the conventional method using pattern matching, there is a limit to templates prepared, and it becomes difficult to accurately detect a tire. There was a case.

The present invention solves such a conventional problem, and a tire detecting method for detecting the presence / absence, size, and position of a tire by using image processing on an input image without installing a treadle. And devices.

[0006]

SUMMARY OF THE INVENTION In order to achieve the above object, the present invention provides an image pickup apparatus for picking up an area around a tire of a passing vehicle from an obliquely upper position by using an image pickup device, and detecting an edge of the obtained image to obtain a tire. The contour information is detected, and a pseudo-Hough transform process for detecting an ellipse with respect to the detected contour information of the tire is defined. At an edge point of the edge detection image, a voting process is performed in a parameter space of the Hough transform. The tire is detected.

Further, the present invention uses the fact that the lower part of the image is a road surface in the input image, takes the road surface as a texture, cuts out a rectangular area on the road surface, defines it as a key texture, and defines its characteristic quantity. Calculate the average value, contrast, and variance value, extract a rectangle from the lower left of the image, calculate the feature amount of the rectangle, and compare the feature amount with the key texture to determine whether the texture is the same, that is, the road surface Is determined, the road surface area is extracted by detecting from the bottom to the top of the image, then the boundary line of the object in contact with the road surface is detected, and the detected boundary line is represented by a quadratic curve. It is characterized in that tire detection is performed by approximation.

Further, in the present invention, in the analysis of the texture feature amount used in the determination of the extraction of the road surface area, it is determined whether the image is a road surface area by using the similarity of the cumulative histogram of the rectangular area, and the image is displayed from bottom to top. The boundary between the road surface and the non-road surface is detected by scanning to the road surface, the boundary line of the object in contact with the road surface region is detected by extracting the road surface region, and the detected boundary line is approximated by a quadratic curve to obtain the tire. Is detected.

Further, the present invention takes advantage of the fact that the density of the tire is lower than the density of the road surface in the input image obtained by imaging the vicinity of the tire of the passing vehicle from an obliquely upper side with an imaging device and taking the image. The road surface area is erased in white by performing a gradation correction that saturates a high-density part of the look-up table, which is a luminance conversion table before and after the conversion used when performing the tone correction, to white. A tire is detected by detecting a boundary line of an existing object and approximating the detected boundary line with a quadratic curve.

Further, the present invention provides a tire detecting apparatus for detecting a tire using the above-described tire detecting method.

Further, the present invention provides a vehicle type discriminating device characterized in that the vehicle type is discriminated using the above-described tire detecting device.

Further, the present invention is characterized in that a fee is settled for a vehicle type determined by using the above-described vehicle type determination device.

[0013]

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

FIG. 1 is a schematic diagram showing the overall configuration of a tire detecting device according to the present invention. An imaging device 1 that images a tire of a passing vehicle traveling on a road from the side, a tire detection device 2 that determines the presence / absence, position, and size of the tire, and a type of the passing vehicle based on information detected by the tire detection device 2 And a vehicle type determination device 3 for performing determination.

The imaging device 1 always captures an image of the roadside space, and the tire detection device 2 creates a background image based on the image and detects the entry of the vehicle based on a difference image from the input image. ing.

The tire detecting device 2 transmits the tire detection information to the vehicle type discriminating device 3, and the vehicle type discriminating device 3 determines the number of axes of the passing vehicle, the size of the tire, etc. based on the information from the tire detecting device 2. Judgment is made comprehensively and finally the vehicle type is judged.

FIG. 2 is a block diagram showing the configuration of the tire detecting device. The tire detection device 2 includes an input control unit 21 that controls the imaging device 1, a power supply unit 22, an overall control unit 23 that controls the entire tire detection device and transmits a tire detection result to the vehicle type determination device 3, An image processing unit 24 that performs a detection image process and an image storage unit 25 that stores an image are connected to the imaging device 1 and the vehicle type determination device 3.

The input control unit 21 converts the captured image into A
/ D conversion, and stores the result in the image storage unit 25 via the overall control unit 23. The image processing unit 24 performs image processing for tire detection on the image stored in the image storage unit 25.

(First Embodiment) FIG. 3 is a flowchart of the tire detection process in the first embodiment based on the pseudo Hough transform. The flow of the tire detection process performed by the image processing unit 24 in FIG. It is. That is, FIG. 3 is a flowchart for explaining a tire detection method using pseudo Hough transform for detecting the shape of a tire.

In an image input step (hereinafter abbreviated as S) 1, an image picked up by the image pickup device 1 is input, and a field correction process S2 is performed. The field correction process S2 is performed for the vehicle moving in the lateral direction with respect to the imaging device 1.
NTSC (National Television Systems Committee)
Abbreviation of method. A broadcasting standard currently used for color broadcasting in Japan and the United States. The NTSC system automatically corrects horizontal displacements in even and odd fields of an image (displayed in an interlaced (interlaced scan) system) using 525 scanning lines in the vertical direction of one screen. This is the process of combining.

FIG. 4 shows the field correction processing S shown in FIG.
FIG. 2 is a diagram for explaining FIG. Since the passing vehicle moves in the lateral direction, that is, in the X direction in terms of the coordinate axes in FIG.
The displacement of each field can be limited in the horizontal direction, that is, in the X direction. Therefore, as shown in FIG. 4A, a region having even and odd fields is cut out, and the cut-out rectangular region of the odd field is moved in the horizontal direction, and the respective cut regions are correlated. As shown in FIG. 4, the shift amount W at which the correlation value becomes the maximum is obtained, and the frame image can be synthesized by adjusting the shift amount W between the even field and the odd field as shown in FIG. Since the shift amount W can be automatically calculated, a frame image can be created by performing field correction automatically.

FIG. 5 shows image images from the field correction processing of S2 shown in FIG. 2 to the tire detection Hough conversion processing and tire recognition processing of S7 and S8. That is, FIG. 5A is a frame composite image obtained by performing the field correction processing S2, FIG. 5B is a processed image in the noise removal S3, FIG. 5C is a processed image in the edge detection S4, and FIG. 5 (d) is a binarization process S
5, (e) is a processed image in the contraction processing S6, and (f) is a tire detection Hough transform S7.
And a processed image in the tire recognition processing S8.

In FIG. 3 and FIG.
3 performs a smoothing process to eliminate minute noise on the road surface. For example, a filter having a variance value intensity of 10 of a Gaussian filter is used.

3 and 5C, the edge detection S
4 is a process for detecting the contour edge of the tire. Since the edge of the tire cannot be detected if the filter size is too small or too large, for example, a 5 × 5 edge detection filter is used. Since the edge of the tire exists on an ellipse, the edge includes all directional components, so that the noise removal processing is not performed according to the direction of the edge.

In FIG. 3 and FIG. 5D, the binarization process S
At 5, the image is binarized to clarify the edge information of the tire. 3 and 5 (e), the contraction processing S6
To remove minute noise on the road surface.

In FIG. 3 and FIG. 5 (f), the tire detection pseudo Hough transform S7 defines the shape of the tire as an ellipse and performs a tire detection pseudo Hough transform process. Originally, the Hough transform is a process of detecting a line, and utilizes the property that when a straight line in an image is mapped to its parameter space using two parameters of inclination and axis intercept, it becomes a point, and the edge on the image is used. By voting for all possible line candidates from points, that is, point candidates in the parameter space, and analyzing the frequency in the parameter space, the straight lines included in the image are extracted. Please refer to "Image Processing Standard Textbook" issued by the Corporate Image Processing Information Education Promotion Association).

By extending the voting condition to an arbitrary figure instead of a line, an arbitrary figure can be detected. In other words, in the case of tire detection, it is assumed that the shape of the tire is an ellipse,
An ellipse is defined using four parameters of center coordinates (X, Y), a radius r, and an oblateness α, and a voting process is performed on all parameters that make the edge point an ellipse with respect to an edge point of an image. The parameter of the number of votes is obtained as an optimal parameter.

FIG. 6 is a diagram for explaining the elliptic Hough transform. As shown in FIG. 6A, the center coordinates (X, Y),
An ellipse is defined by four parameters of a radius r and an oblateness α (= a2 / a1). When a1 = a2, the flattening rate α = 1, and
The ellipse becomes a circle.

If the Hough transform is performed as it is using the four parameters, a four-dimensional voting table is required. Therefore, as shown in FIG. 6B, if the flatness α and the radius r are constant values, It becomes a two-dimensional table of the center coordinates (X, Y).

As shown in FIG. 6C, the center coordinates (X,
A voting process is performed on Y). The candidate point (X, Y) of the center coordinates is calculated by the following equation (1). In Equation (1), θ is the rotation angle, and usually may be changed by 1 degree.

[0031]

(Equation 1)

When the voting process is performed on the center coordinate candidates for all the edge detection points, the number of votes is concentrated around the true value of the center coordinates as shown in FIG. For example, the center coordinates (X, Y) are determined.

As shown in FIG. 6 (e), the radius r is changed, voting is again performed for the center coordinate candidate, and the maximum voting number is checked. If the radius r is a true value, the center coordinates that are the true values are included in the candidate points, and the number of votes becomes the maximum value. That is,
If the radius r is not the true value, as shown in FIG. 6F, the peak of the vote count is lower than when the radius r is the true value.

In FIG. 3 and FIG. 5 (f), a tire recognition process S8 analyzes the voting table created by the tire detection Hough transform S7 to determine the presence / absence, position, and size of the tire. Since four parameters are determined by the tire detection Hough transform, the ellipse is determined, that is, the position of the tire is determined.

By calculating the variance value of the voting table, as shown in FIG. 7 (a), when there is a tire, the voting table has a peak and the variance value of the voting table decreases. When there is no tire as shown in FIG. 7 (b), the voting value table does not have a peak and becomes smooth as a whole, so that the variance value of the voting table becomes large.

Also, without using the maximum voting value of the voting table, an area having a certain number of votes or more is taken out near the maximum value and the center of gravity is set to the center coordinates (X, Y), so that the influence of noise is reduced. The center coordinates can be calculated without the need. In the above description, the flatness α is a constant value. However, by using a variable value, more accurate tire detection can be performed.

(Second Embodiment) FIG. 8 is a flowchart for explaining a process when tire detection is performed in the second embodiment based on texture features.
Is performed by the image processing unit 24. The texture means a pattern represented by regular fine gradation changes such as wood grain. In a road surface region of an image, a region of a certain size (for example, 8 × 8 pixels) is cut out, and the region is regarded as a texture, and the feature amount (average value, contrast,
Variance) to extract the same texture,
That is, this is a process of extracting a road surface region and linking it to detection of a tire in contact with the road surface.

As shown in FIG. 8, first, an image is input in image input S9 (step 9), and a field correction process S1 is performed.
Perform 0. The above-described field correction processing S2 and field correction processing S10 are the same processing.

FIG. 9 shows an image image of a tire detection method based on a texture feature. As shown in FIG. 9A, an area with a road surface is cut out in key texture setting S11, set as a texture serving as a key, and defined as original data for comparing road surface feature amounts.
Then, an average value, a contrast, and a variance value are calculated as texture feature amounts (S12).

Next, as shown in FIG. 9B, an area having the same size as the key texture is cut out in the lower left part of the image, and the feature amount is calculated similarly. By comparing the feature amounts of the key texture with these feature amounts, if they are the same, it is possible to determine that the area is a road surface, proceed upward in the image, cut out a new rectangle, and compare the features. Is repeatedly performed (S13).

When it is determined that the road is not the road surface C (= constant) times, the road surface search processing is terminated, and a new road surface search is performed from below the image of the next line as shown in FIG. 9C. Go. The above road surface search flow is shown in FIG. FIG. 9D shows a state in which the road surface search processing has been performed up to the right end of the image.

In texture noise removal S14, FIG.
As shown in (e), the detected road surface area is subjected to erosion processing after expansion processing to eliminate singular points in the road surface area and remove noise.

The road surface area detected in the road surface recognition processing S15 is finally determined. As shown in FIG. 9F, the boundary line of the object in contact with the road surface is detected in tire recognition processing S16, and approximation is performed using a quadratic curve. If the quadratic coefficient is large to some extent, it can be determined that the object has an arc, and the presence or absence of a tire can be determined.

The quadratic curve approximation at the edge detection point of the tire shown in FIG. Edge detection point P1 (x ₁ ,
For n data of y ₁ ), P ₂ (x ₂ , y ₂ ),..., Pn (x _n , y _n ), the quadratic curve L1 approximates the equation of Y = ax ² + bx + c.

This is solved by the least squares method and summarized as follows.
Since the following equation, that is, the simultaneous linear equation shown in Expression 2, is solved for a, b, and c, it can be approximated by a quadratic curve.

[0046]

(Equation 2)

The lowest point of the quadratic curve is the lowest point of the tire, and as shown in FIG. 9 (g), the density of the tire can be limited to some extent. By performing (project), the position of the tire (horizontal H × vertical V) can be accurately obtained, and the detection result is shown in FIG. 9 (h).

(Third Embodiment) FIG. 11 is a flow chart for explaining the processing when tire detection is performed in the third embodiment based on a rectangular cumulative histogram. Perform at 24. In FIG. 11, an image is input by image input S17 (step 17), and field correction processing S18 is performed. Since the above-described field correction processing S2 and field correction processing S18 are the same processing, the description thereof will be omitted.

FIG. 12 is an image image for explaining a tire detection method based on a rectangular cumulative histogram. As shown in FIG. 12A, the road surface search processing (lower) S
In step 19, an area of 4 × 4 or 8 × 8 pixels is cut out from the lower left of the image, and the cumulative histogram K1 is obtained.

Next, an area of the same size is cut out thereon,
The cumulative histogram K2 is obtained. The frequency values H1 and H2 of the respective densities are compared, and if the absolute value of the difference is equal to or smaller than a constant value, it can be said that the shapes of the cumulative histograms are similar, that is, the same road surface.

If it is a road surface, as shown in FIG. 12 (b), the area immediately above is cut out and the shapes of the cumulative histograms are sequentially compared. The process is terminated where the shapes are different, and as shown in FIG. 12 (c), the road surface is searched upward in the same right line. FIG. 12D shows the result of finally performing the search to the right end of the screen and performing the road surface search. The processing flow so far is exactly the same as the flow for searching for a road surface based on the texture feature according to the above-described second embodiment.

Next, the process of performing a road surface search in the right-to-left and left-to-right directions of the image is a road surface search process (left and right) S20, but the processing flow itself is exactly the same as that of the road surface search (lower) S19. .

In the road surface noise removal S21, the singular point noise is removed in the contraction / expansion processing, but the processing content is the same as the texture noise removal according to the above-described second embodiment.

As shown in FIG. 12 (e), the road surface recognition process S22 includes a search area for the road surface from the bottom of the image, a road surface search area from the right side of the image, and a road surface search area from the left side of the image. And finally determine the road surface area. The tire recognition processing S23 is the same processing as the tire recognition processing S16 according to the above-described second embodiment, and can similarly detect the presence / absence, size, and position of a tire.

(Fourth Embodiment) FIG. 13 shows a process for performing tire detection in the fourth embodiment by saturating a high density portion of a look-up table with white and erasing a road surface region with white. It is a flowchart, FIG.
Is performed by the image processing unit 24. In FIG. 13, image input S2
4 (Step 24), an image is input and the input image S2
4 is subjected to field correction processing S25. The above-described field correction processing S2 and field correction processing S25
Is the same process.

The noise reduction S26 removes minute noise on the road surface, and the lookup table correction S27 adjusts the lookup table as shown in FIG. 14 (gradation correction) to saturate the road surface area white. As shown in FIG. 15, only a tire region with low luminance is extracted.

The road surface recognition processing S28 is the same as the road surface recognition processing S15 in the above-described second embodiment, and the tire recognition processing S27 is the same as the tire recognition processing S16 in the above-described second embodiment. , Size and position can be detected.

As described above, four types of tire detection methods have been described as shown in the first to fourth embodiments. All or some of these detection methods are combined to form an AND condition or an OR condition. Of course, it is also possible to perform tire detection.

Further, it is possible to provide a tire detecting apparatus for detecting a tire at a tollgate on an expressway by using the above-described tire detecting method. , A vehicle type discriminating apparatus can be provided.

In a location where the tread is difficult to install due to its structure, and where it is impossible to measure the number of axes with the conventional tread, it is also possible to collect a fee for each type of vehicle discriminated by the type of vehicle discrimination device. It is also possible to provide a device.

As described above, the image of the vicinity of the tire of the passing vehicle is taken obliquely from above by the image pickup device, and recognition processing is performed using image processing on the basis of the features of the tire and the feature amount of the road surface area. Since it is possible to detect the fact and to measure the exact size and the position of the tire, it is possible to collect an accurate fee.

[0062]

As described above, according to the present invention, an image is obtained based on the characteristics of the tire and the feature amount of the road surface area by using an input image obtained from a camera which captures the tire from a diagonally upper side without installing a tread plate. The recognition processing is performed using the processing, the effect that the tire is easily detected can be obtained, and the accurate size and the position of the tire can be measured.

[Brief description of the drawings]

FIG. 1 is a schematic diagram showing the overall configuration of a tire detection device according to the present invention;

FIG. 2 is a block diagram showing a configuration of a tire detection device according to the present invention.

FIG. 3 is a flowchart of a tire detection process according to the first embodiment based on a pseudo Hough transform;

FIG. 4 is a view for explaining a field correction process S2 shown in FIG. 3;

FIG. 5 is a flow chart showing the processing from the field correction processing of S2 shown in FIG. 2 to S
The figure which shows the image image until it reaches the tire detection Hough conversion process of 7 and S8, and the tire recognition process.

FIG. 6 is a diagram for explaining an elliptic Hough transform.

FIG. 7 is a diagram for explaining determination of the presence or absence of a tire based on a voting table;

FIG. 8 is a flowchart for explaining processing when tire detection is performed in the second embodiment based on texture features;

FIG. 9 is a diagram showing an image image of a tire detection method based on a texture feature;

FIG. 10 is a flowchart of a road surface search,

FIG. 11 is a flowchart for explaining processing when tire detection is performed in the third embodiment based on a rectangular cumulative histogram;

FIG. 12 is a diagram showing an image image for explaining a tire detection method based on a rectangular cumulative histogram;

FIG. 13 is a flowchart of a process of performing tire detection according to the fourth embodiment by saturating a high-density portion of a lookup table with white and erasing a road surface region with white;

FIG. 14 is a diagram illustrating an example of a look-up table that saturates a road surface with white;

FIG. 15 is a conversion example using a look-up table that saturates a road surface with white.

[Explanation of symbols]

 REFERENCE SIGNS LIST 1 imaging device 2 tire detection device 3 vehicle type discrimination device 21 input control unit 22 power supply unit 23 overall control unit 24 image processing unit 25 image storage unit

──────────────────────────────────────────────────続 き Continued on the front page (51) Int.Cl. ⁷ Identification symbol FI Theme coat ゛ (Reference) G08G 1/04 G01B 11/24 K F Term (Reference) 2F065 AA51 AA61 CC13 DD02 DD07 FF04 HH12 QQ03 QQ24 QQ31 QQ32 QQ33 QQ34 QQ42 QQ43 SS01 UU05 5H180 CC04 EE07 EE10 5L096 AA02 AA06 BA04 CA02 DA02 EA12 FA06 FA32 FA33 FA35 JA03

Claims (7)

[Claims] 1. An image pickup device captures the vicinity of a tire of a passing vehicle from an obliquely upper side with an image pickup device, detects an edge of the obtained image, detects tire contour information, and detects an ellipse with respect to the detected tire contour information. A quasi-Hough transform process for detecting a tire, and voting in an Hough transform parameter space to detect a tire at an edge point of an edge detection image. 2. An input image obtained by imaging the vicinity of a tire of a passing vehicle from an obliquely upper side with an imaging device and capturing the image,
Using that the lower part of the image is the road surface, the road surface is regarded as texture, a rectangular area on the road surface is cut out and defined as the key texture, and the average value, contrast, variance value that is the feature amount is calculated, In addition, a rectangle is cut out from the lower left of the image, the feature amount of the rectangle is calculated, the feature amount is compared with the key texture to determine whether the texture is the same, and the image is detected from the bottom to the top. A road surface area, thereby detecting a boundary line of an object in contact with the road surface, and detecting the tire by approximating the detected boundary line with a quadratic curve. 3. In the analysis of the texture feature amount used in the determination of the extraction of the road surface area, it is determined whether the road area is a road area using the similarity of the cumulative histogram of the rectangular area, and the image is scanned from bottom to top. Detecting the boundary between the road surface and the non-road surface by moving, detecting the boundary line of the object in contact with the road surface region by extracting the road surface region, and detecting the tire by approximating the detected boundary line with a quadratic curve. 3. The tire detection method according to claim 2, wherein: 4. An input image obtained by imaging the vicinity of a tire of a passing vehicle from diagonally above from above with an imaging device and capturing the image,
Taking advantage of the fact that the density of the tire is lower than the density of the road surface, gradation correction that saturates the high-density part of the lookup table, which is the luminance conversion table before and after conversion used for gradation correction, to white. To erase the road surface area in white, detect the boundary line of the object in contact with the road surface,
A tire detection method, wherein tire detection is performed by approximating a detected boundary line with a quadratic curve. 5. A tire detection device for detecting a tire using the tire detection method according to claim 1. 6. A vehicle type discriminating apparatus for discriminating a vehicle type using the tire detecting device according to claim 5. 7. A toll settlement method, wherein a toll settlement is performed for a vehicle type determined by using the vehicle type identification device according to claim 6.