CN108399412A - A kind of X-type angular-point sub-pixel extracting method - Google Patents

Abstract

The invention proposes an X-shaped corner point sub-pixel extraction method, which belongs to the field of image processing technology. The method uses the pixel point in the upper left corner of the grayscale image as the origin to establish a coordinate system. The horizontal direction of the grayscale image is the X axis, and the vertical direction is the X axis. The direction is the Y axis, and the Harris operator is used to detect the corners of the gray image; the coordinates of the corners are determined by the Harris operator, and the sub-pixel-level corner coordinates are obtained according to the coordinates of the corners; The point is the center of the window and the side length of the window is set, the window is constructed, and the corner point with the largest amount of corner points in the window is retained, and other corner points are deleted; this method can filter out the false corner points in the image, and can accurately detect the corner points in the image points, the detection accuracy can reach 0.09 pixels, this method is simple to calculate, and is insensitive to image rotation, gray scale change, noise influence and viewpoint change.

Description

A method for sub-pixel extraction of X-shaped corner points

technical field

The invention belongs to the technical field of image processing, and in particular relates to an X-shaped corner point sub-pixel extraction method.

Background technique

With the continuous development of science and technology, cameras have also developed rapidly, and there are more and more types and functions of cameras, and people have also enjoyed various conveniences brought by the development of science and technology. For example, people can obtain the environmental information in the parking lot by setting the camera on the vehicle and collecting images through the camera.

The current cameras need to calibrate the parameters to obtain more accurate parameters, and the calculation of parameters is directly affected by the accuracy of the corner points, which requires a higher accuracy of corner point extraction; traditional corner point extraction techniques have The corner detection method based on the segment test, but the image scene is complex, and the template size of the segment test, the threshold of the gray difference is difficult to choose, etc., resulting in the detection of X-shaped corner points, it is easy to cause missed detection, multiple detection, There are more false corners.

Contents of the invention

Aiming at the deficiencies of the prior art, the present invention proposes an X-shaped corner point sub-pixel extraction method, which can accurately identify the X-shaped corner point in the image, and provides guarantee for accurate calibration of the camera.

A method for extracting X-shaped corner sub-pixels, comprising the following steps:

Step 1, converting the image to be processed into a grayscale image;

Step 2, establish a coordinate system with the pixel point in the upper left corner of the grayscale image as the origin, the horizontal direction of the grayscale image is the X axis, the vertical direction is the Y axis, and the Harris operator is used to detect the corners of the grayscale image;

Step 3, determine the coordinate position of the corner point through the Harris operator, and obtain the sub-pixel level corner point coordinates according to the coordinate position of the corner point;

Step 4. Take each sub-pixel-level corner point as the center of the window and set the window side length to construct the window, retain the corner point with the largest amount of corner points in the window, and delete other corner points;

Step 5, judging whether the number of corner points of the image under test is known, if so, then perform step 6, otherwise perform step 7;

Step 6. Sorting the retained corner points according to the amount of corner points from large to small, and selecting the corner point in front according to the number of known corner points in the image to complete the extraction of the X-shaped corner point sub-pixels of the image;

Step 7. Judging whether the amount of corner points of the retained corner points is greater than the set threshold, if so, keep them, and complete the extraction of the X-shaped corner point sub-pixels of the image; otherwise, delete them.

For the X-shaped corner point, if any point around the X-shaped corner point p is q, then the dot product of the gradient at point p and the vector qp is zero.

Advantages of the present invention:

The present invention proposes an X-shaped corner point sub-pixel extraction method, which can filter out false corner points in the image, and can accurately detect the corner points in the image, and the detection accuracy can reach 0.09 pixels. Insensitive to image rotation, grayscale changes, noise effects and viewpoint changes.

Description of drawings

Fig. 1 is a flow chart of an X-shaped corner point sub-pixel extraction method according to an embodiment of the present invention;

Fig. 2 is the image to be processed of an embodiment of the present invention;

Fig. 3 is the grayscale image of the image to be processed in an embodiment of the present invention;

Fig. 4 is a Gaussian image of an embodiment of the present invention;

Fig. 5 is the filtered and corroded grayscale image of an embodiment of the present invention;

FIG. 6 is a schematic diagram of pixel-level coordinates extracted by using Harris operator according to an embodiment of the present invention;

7 is a schematic diagram of sub-pixel-level coordinates extracted by a Harris operator according to an embodiment of the present invention;

Fig. 8 is a schematic diagram of a sub-pixel level X-shaped corner point vector according to an embodiment of the present invention;

FIG. 9 is a schematic diagram of corner point coordinates after deletion of pixel-level corner points according to an embodiment of the present invention;

FIG. 10 is a schematic diagram of corner point coordinates after deletion of sub-pixel level corner points according to an embodiment of the present invention;

Fig. 11 is a schematic diagram of sub-pixel level X-shaped corner points when the number of corner points is known in an embodiment of the present invention;

Fig. 12 is a schematic diagram of sub-pixel level X-shaped corner points when the number of corner points is unknown according to an embodiment of the present invention;

Fig. 13 is a schematic diagram of an extraction result of corner points based on pixel level according to an embodiment of the present invention.

Detailed ways

An embodiment of the present invention will be further described below in conjunction with the accompanying drawings.

In the embodiment of the present invention, the X-shaped corner point sub-pixel extraction method, the flow chart of the method is shown in Figure 1, including the following steps:

Step 1, converting the image to be processed into a grayscale image;

In the embodiment of the present invention, the Harris corner point is a point on the image whose gray value changes relatively large in any two vertical directions. However, the Harris algorithm extracts the corner points from the grayscale image, so the original image is first converted into a grayscale image. Figure 2 is the original image, and Figure 3 is the corresponding grayscale image;

Step 2, establish a coordinate system with the point in the upper left corner of the grayscale image as the origin, the horizontal direction of the grayscale image is the X axis, the vertical direction is the Y axis, and the Harris operator is used to detect the corners of the grayscale image;

In the embodiment of the present invention, a coordinate system is established with the point in the upper left corner of the image as the origin, wherein the horizontal direction of the image is the X axis, and the vertical direction of the image is the Y axis, and the corner point detection is performed through the Harris operator. The specific detection process is as follows :

Step 2-1, using the gradient operator to calculate the difference in x and y directions;

Step 2-2, using the grayscale of each pixel in the image to calculate the gradients I _x and I _y in the x-axis and y-axis directions of each pixel in the image;

Calculated as follows:

Wherein, in formula (1) and formula (2), the gradient operator can be the operator shown in table 1;

Table 1

Step 2-3, respectively calculating the gradient products I _x ² , I _y ² and I _xy of each pixel point in the image in two directions;

Calculated as follows:

I _x ² =I _x *I _x (3)

I _y ² =I _y *I _y (4)

I _xy =I _x *I _y (5)

Step 2-4, use the Gaussian function to generate a Gaussian kernel to filter I _x ² , I _y ² and I _xy respectively, and the Gaussian filter image is shown in Figure 4;

The Gaussian function formula is:

Among them, x represents the horizontal direction, y represents the vertical direction, σ is the standard deviation of the two-dimensional Gaussian curve, g represents the convolution, and W is the Gaussian filter window.

According to formula (6) to formula (9), calculate the corner value R of each pixel point, the calculation formula is as follows:

In the embodiment of the present invention, the corner point corresponding to the local maximum value of the corner point amount R is found from the obtained corner point amount R, which is the X-shaped corner point to be extracted; It is not the points that need to be extracted, so in the embodiment of the present invention, the shape similar to a checkerboard is used to perform median filtering and erosion operations on the obtained grayscale image, so that the edges and corners of the image shape are not obvious, reducing We do not expect the R value of the corners to weaken the corner features of such corners, and at the same time avoid the blur caused by the X-shaped corners that should be filtered, so that the R value of the X-shaped corners in the image is the highest; Figure 5 is after Grayscale image after filtering and erosion operations.

Step 3, determine the coordinate position of the corner point through the Harris operator, and obtain the sub-pixel level corner point coordinates according to the coordinate position of the corner point;

In the embodiment of the present invention, the Harris operator is used to determine the coordinate position of the corner point, and the sub-pixel-level corner point coordinates are calculated according to the coordinate position of the corner point, wherein, an X-shaped corner point p is set, and any point around p is set to q, then The dot product of the gradient at point p and the vector qP is zero; the pixel-level coordinates extracted by the Harris operator are shown in Figure 6, and the sub-pixel-level coordinates are shown in Figure 7;

In the embodiment of the present invention, it can be seen from Fig. 7 that there are and only 7 Harris corner points can be detected in the black and white grid area paved with large white paper, and the detected positions are relatively accurate, and one of the 7 corner points is the required markers. Due to the relatively complex scene, many non-marked points are also detected in Figure 6, such as the points located outside the scene in Figure 6, that is, the big white paper, and the position coordinates of the corner points in Figure 6 are only at the pixel level, and further purification is needed to obtain sub- pixel level;

In the embodiment of the present invention, since the dot product of the gradient at the point p and the vector qp is zero, according to this theorem, the vector shown in Figure 8 can be constructed;

In the embodiment of the present invention, the positions of points near the corner area can be divided into two types: on the edge and not on the edge. In Figure 8, assuming that point B is on the edge, its gray gradient vector perpendicular to the vector OB, then The dot product with vector OB is 0; point A is not on the edge, since the gray gradient vector of this point is 0, so The dot product with the vector OA is also 0, which can be expressed by the following formula:

in, Represents the gray gradient vector, Represents the vector from the origin of the image to the coordinate point i, Represents the vector from the origin of the image to the coordinate point O;

In the embodiment of the present invention, according to the formula (11), multiple points can be found near point O, and the gray gradient vector of each point of the multiple points that are found can be calculated as well as Composing a system of equations can solve the system of equations to obtain the coordinate value of the sub-pixel level precision of the corner point O to complete the extraction of the sub-pixel of the corner point in the image. The sub-pixel level coordinates are shown in Figure 7.

Step 4. Take each sub-pixel-level corner point as the center of the window and set the window side length to construct the window, retain the corner point with the largest amount of corner points in the window, and delete other corner points;

In the embodiment of the present invention, for the found corner points, corner points often gather, which can also be clearly seen in Figure 6 and Figure 7, and this type of corner point is called a repeated corner point, which needs to Select a corner point with the largest R value at the place where the corner points gather; the specific method is as follows:

Take each found corner point as the center of the window, set the size as an n×n window, find the corner point corresponding to the maximum value of the corner point value of each pixel in the n×n window, and delete other corner points at the same time , select a suitable n value according to the specific image size, n is 5 in the embodiment of the present invention, and the unit is a pixel; Figure 9 is the result after deletion in Figure 6, and Figure 10 is the result after deletion in Figure 7;

According to the actual corner position on the white paper, the accuracy of the corner coordinates extracted by using sub-pixels has been greatly improved, and it is closer to the actual corner point, and the corner position is more accurate.

Step 5, judging whether the number of corner points of the image under test is known, if so, then perform step 6, otherwise perform step 7;

Step 6. Sorting the retained corner points according to the amount of corner points from large to small, and selecting the corner point in front according to the number of known corner points in the image to complete the extraction of the X-shaped corner point sub-pixels of the image;

In the embodiment of the present invention, the corner point is obtained by calculating the R value of the corner point, and the R value of the X-shaped corner point is larger than the R value of the corner point on the contour, so when the number N of corner points in the known image is At this time, all the corner points in the image are sorted according to the value of R, and the first N corner points are selected as the X-shaped corner points; in the embodiment of the present invention, 4 sheets of white paper are laid on the floor, so N is set to 4, namely It is known that there are 4 X-shaped corner points;

Step 7. Judging whether the amount of corner points of the retained corner points is greater than the set threshold, if so, keep them, and complete the extraction of the X-shaped corner point sub-pixels of the image; otherwise, delete them.

In the embodiment of the present invention, a threshold is set when using the Harris operator to extract corner points. When the R value of the point is greater than the threshold, it is regarded as a corner point. When the R value of the point is smaller than the threshold, it is not a corner. point. For X-shaped corners, their R values are generally too large. Therefore, in the embodiment of the present invention, when setting the threshold, the threshold can be appropriately increased to extract only X-shaped corners; the selection of the threshold should be done without missing The X-shaped corner points cannot be extracted, and the non-X-shaped corner points cannot be extracted.

Experiments have been carried out for the above two methods. Figure 11 is the result obtained with the known number of corner points, and Figure 12 is the result obtained with the unknown number of corner points; it can be seen from the results that the X-type extracted by the two methods The corner points are consistent and in line with the actual situation, and have a good effect on the extraction of X-shaped corner points in complex backgrounds; similarly, the above two methods are used to extract pixel-level corner points, and the same results are obtained, as shown in Figure 13 It is based on the pixel-level corner point extraction results; comparing the two, it is found that the sub-pixel-level X-shaped corner point position is more accurate and more in line with the actual situation.

Claims (2)

1. A method for extracting X-type corner sub-pixels is characterized in that, comprising the following steps: Step 1, converting the image to be processed into a grayscale image; Step 2, establish a coordinate system with the pixel point in the upper left corner of the grayscale image as the origin, the horizontal direction of the grayscale image is the X axis, the vertical direction is the Y axis, and the Harris operator is used to detect the corners of the grayscale image; Step 3, determine the coordinate position of the corner point through the Harris operator, and obtain the sub-pixel level corner point coordinates according to the coordinate position of the corner point; Step 4. Take each sub-pixel-level corner point as the center of the window and set the window side length to construct the window, retain the corner point with the largest amount of corner points in the window, and delete other corner points; Step 5, judging whether the number of corner points of the image under test is known, if so, then perform step 6, otherwise perform step 7; Step 6. Sorting the retained corner points according to the amount of corner points from large to small, and selecting the corner point in front according to the number of known corner points in the image to complete the extraction of the X-shaped corner point sub-pixels of the image; Step 7. Judging whether the amount of corner points of the retained corner points is greater than the set threshold, if so, keep them, and complete the extraction of the X-shaped corner point sub-pixels of the image; otherwise, delete them. 2. X-type corner point sub-pixel extraction method according to claim 1, is characterized in that, described X-type corner point, if any point around X-type corner point p is q, then the gradient and vector at p point place The qp dot product is zero.