METHOD OF CAPTURING AND RECONSTRUCTING COURT LINES

Abstract

A method of extracting and reconstructing court lines includes the steps of binarizing a court image of a court including court lines to form a binary image; performing horizontal projection for the binary image; searching for plural corners in the binary image and defining a court line range by the corners; forming plural linear segments from images within the court line range by linear transformation; defining at least one first cluster and at least one second cluster according to the characteristics of the linear segments and categorizing the linear segments into plural groups; taking an average of each group as a standard court line and creating a linear equation of the standard court line to locate the point of intersection of the standard court lines; and reconstructing the court lines according to the point of intersection. This method is capable of extracting the image of a portion of the court line from a dynamic or static image having a court line quickly to eliminate interference caused by noises coming from a portion other than the court line such as the background color, ambient brightness, people or advertisement, and reconstructing the court lines quickly and accurately to facilitate the determination of the boundary of a court line or the computation of data.

Claims

1. A method of extracting and reconstructing court lines, comprising the steps of: binarizing a court image of a court including a court line to form a binary image; searching for a plurality of corners in the binary image and defining a court line range by the corners; forming a plurality of linear segments from an image within the court line range by linear transformation; defining at least one first cluster and at least one second cluster according to the characteristics of the linear segments, and categorizing the linear segments into a plurality of groups according to the first cluster and the second cluster; taking an average of each group as a standard court line, and creating a linear equation of the standard court line to locate the position of a point of intersection of the standard court lines; and reconstructing the court line according to the point of intersection.

2. The method of extracting and reconstructing court lines according to claim 1, further comprising the sub-steps of performing a gradient computation of the court image to produce a horizontal gradient image and a vertical gradient image, and combining the horizontal gradient image and the vertical gradient image to form the binary image.

3. The method of extracting and reconstructing court lines according to claim 2, further comprising the sub-steps of defining a threshold according to the color of the court line, and forming the binary image by the threshold screening when the horizontal gradient image and the vertical gradient image are combined.

4. The method of extracting and reconstructing court lines according to claim 1, further comprising the sub-steps of: performing a horizontal projection of the binary image to form a first horizontal projection image; defining a range of the first horizontal projection image with a horizontal cumulative value greater than a cumulative threshold to be a search range, and searching for the corners in the search range.

5. The method of extracting and reconstructing court lines according to claim 4, further comprising the sub-steps of: using Equation 1 to filter out the noise of the first horizontal projection image to form a second horizontal projection image: $\begin{matrix} F_{i} = {\begin{matrix} \frac{.Math. p_{i} - μ .Math. \times .Math. k}{σ}, p_{i} > μ \\ 0, Otherwise \end{matrix}; & (Equation .Math. .Math. 1) \end{matrix}$ wherein, F.sub.i is a second horizontal projection image; p.sub.i is a horizontal cumulative value of the corresponding first horizontal projection image; μ is an average of the horizontal cumulative values of the first horizontal projection image; and σ is a standard deviation; defining the cumulative threshold by Equation 2: $\begin{matrix} T = \max (\frac{F_{i}}{ρ}); & (Equation .Math. .Math. 2) \end{matrix}$ wherein, ρ is a magnification constant; and defining a range of the second horizontal projection image with a horizontal cumulative value greater than the cumulative threshold as the search range.

6. The method of extracting and reconstructing court lines according to claim 4, wherein the first horizontal projection image is formed by performing a horizontal projection at a middle third of the binary image.

7. The method of extracting and reconstructing court lines according to claim 1, further comprising the sub-steps of: setting the court line as a quadrilateral, and creating a binary search image and its search coordinates, and dividing the search image into an upper left blank, an upper right blank, a lower left blank and a lower right blank through the search coordinates; slidably searching the search image in the binary image, and performing a convolution of the binary image; and defining the maximum after convolution takes place as the corner.

8. The method of extracting and reconstructing court lines according to claim 7, wherein the convolution of the search image and the binary image is carried out by Equation 3:
O(i,j)=Σ.sub.s=−4.sup.4Σ.sub.t=−4.sup.4m(s,t)×p(i+s,j+t);(i*,j*)=argmax O(i,j) (Equation 3) wherein O(i,j) is a corner; m(s,t) is a search image; and p(i,j) is a binary image.

9. The method of extracting and reconstructing court lines according to claim 1, further comprising the sub-steps of performing a thinning process after the image in the court line range is processed by a closing process, and then forming the linear segment by linear transformation.

10. The method of extracting and reconstructing court lines according to claim 1, further comprising a sub-step of performing a Hough transform of the court line in the court line range to form the linear segment.

11. The method of extracting and reconstructing court lines according to claim 1, further comprising the sub-steps of: using the first cluster to classify a horizontal segment in the linear segment according to the slope and the position of the Y-axis coordinate; and using the second cluster to classify a vertical segment in the linear segment according to the slope and the intercept.

12. The method of extracting and reconstructing court lines according to claim 11, wherein the court line is a tennis court line, and six first clusters and five second clusters are defined according to the characteristics of the linear segment.

13. The method of extracting and reconstructing court lines according to claim 1, wherein the point of intersection is used to reconstruct the court line according to the court line position by a line function.

14. The method of extracting and reconstructing court lines according to claim 1, wherein the court line is a tennis court line, and the standard court line is provided for computing 30 points of intersection.

15. The method of extracting and reconstructing court lines according to claim 14, further comprising the sub-steps of: setting the court image as a dynamic continuous image, defining a first constant value, a second constant value and a computing value, computing the distance value between the point of intersection of the current court image position and the point of intersection of the previous court image at the corresponding position, and increasing the computing value if the distance value is smaller than the first constant value, and computing an error threshold T.sub.e by Equation 4 if the computing value is greater than the second constant value: $\begin{matrix} T_{e} = \frac{α}{30} [{.Math.}_{i = 1}^{30} .Math. .Math. .Math. P_{i} (k + 1) - P_{i} (k) .Math.] & (Equation .Math. .Math. 4) \end{matrix}$ wherein α is an error magnification constant; and replacing the reconstructed court line of the current frame by the reconstructed court line of the previous frame if the distance value is greater than the error threshold at any point of intersection.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0027] FIG. 1 is a flow chart of the present invention;

[0028] FIG. 2 is a flow chart of converting a court image into a binary image by the Sobel algorithm in accordance with the present invention;

[0029] FIG. 3 is a schematic view of a horizontal gradient image of the present invention;

[0030] FIG. 4 is a schematic view of a vertical gradient image of the present invention;

[0031] FIG. 5 is a schematic view of a binary image of the present invention;

[0032] FIG. 6 is a schematic view of a first horizontal projection image of the present invention;

[0033] FIG. 7 is a schematic view of a second horizontal projection image of the present invention;

[0034] FIG. 8 is a schematic view of a search image in a binary image of the present invention;

[0035] FIG. 9a is a schematic view of a pixel value of the upper left blank of a search image of the present invention;

[0036] FIG. 9b is a schematic view of a pixel value of the upper right blank of a search image of the present invention;

[0037] FIG. 9c is a schematic view of a pixel value of the lower left blank of a search image of the present invention;

[0038] FIG. 9d is a schematic view of a pixel value of the lower right blank of a search image of the present invention;

[0039] FIG. 10 is a flow chart of the step of performing a closing process to the step of generating a focal point in accordance with the present invention;

[0040] FIG. 11 is a schematic view of the present invention after a closing process takes place;

[0041] FIG. 12 is a schematic view of the present invention after a thinning process takes place;

[0042] FIG. 13 is a schematic view of the present invention after a Hough transform takes place;

[0043] FIG. 14 is a classification chart of a linear segment classified by a first cluster as a different group in accordance with the present invention;

[0044] FIG. 15 is a classification chart of a linear segment classified by a second cluster as a different group in accordance with the present invention;

[0045] FIG. 16 is a schematic view of defining a standard court line in accordance with the present invention;

[0046] FIG. 17 is a schematic view of defining a point of intersection and reconstructing a court line in accordance with the present invention;

[0047] FIGS. 18 and 19 are experiment charts of creating a standard court line in a court image in accordance with the present invention;

[0048] FIGS. 20 and 21 are experiment charts of creating a point of intersection in a court image in accordance with the present invention; and

[0049] FIGS. 22 to 29 are experiment charts of reconstructing a court line in a court image in accordance with the present invention.

DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0050] The technical contents of this disclosure will become apparent with the detailed description of preferred embodiments accompanied with the illustration of related drawings as follows. It is intended that the embodiments and figures disclosed herein are to be considered illustrative rather than restrictive.

[0051] With reference to FIG. 1 for a method of extracting and reconstructing court lines in accordance with the present invention, the method comprises the following steps:

[0052] S001: Capture a court image including a court line 1. It is noteworthy that the court image may include interference caused by a complicated background including audience, advertising signs, referees, a net, and players. To reduce the interference of the site background and the distortion of a video after being compressed and eliminate the possibility of losing a part of the court line which is stepped by the players or worn out, an embodiment of the present invention adopts the Sobel algorithm for gradient computation to obtain a high-quality binary image 2 while converting the court image into the binary image 2. In Steps S101 to S110 as shown in FIG. 2, a gradient computation of the court image is performed to generate a horizontal gradient image 21 and a vertical gradient image 22, a median filtering is performed to reduce noises as shown in FIGS. 3 and 4, and the ratio of the horizontal gradient of the horizontal gradient image 21 to the vertical gradient of the vertical gradient image 22 is increased, and the horizontal gradient image 21 and the vertical gradient image 22 are combined. Since the court lines 1 of the court are of the same color. With respect to the tennis court, the court line 1 is in white color, and a threshold may be defined according to a color range of the court line 1, and the threshold is used for screening the binary image 2 as shown in FIG. 5, and the binary image 2 is in the same color of the court line 1 (which is white in color), and the remaining ones are black in color. The top edge of the net in the tennis court is painted with a white thick line, so that the net image is also white in color.

[0053] S002: Perform a horizontal projection of the binary image 2 to form a first horizontal projection image 3 as shown in FIG. 6, since the original court image still includes a background having the same color of the court line 1. In general, the court line 1 has a plurality of segments in specific directions. As to the tennis court line, there are more horizontal lines than vertical lines, so that the horizontal cumulative value obtained after the horizontal projection is divided into court line 1 and background noise. The horizontal projection of this embodiment refers to the plurality of segments in specific directions in the court line 1, but the present invention is not limited to the horizontal direction.

[0054] Define a cumulative threshold, and define a range of the first horizontal projection image 3 having a horizontal cumulative value greater than the cumulative threshold as a search range 5, and locate a corner in the search range 5.

[0055] In a preferred embodiment, the noises of the first horizontal projection image 3 are filtered by the following Mathematical Equation 1 to form a second horizontal projection image 4 as shown in FIG. 7 in order to distinguish the court line 1 in the first horizontal projection image 3 and the background more easily.

[00004] $\begin{matrix} F_{i} = {\begin{matrix} \frac{.Math. p_{i} - μ .Math. \times .Math. k}{σ}, p_{i} > μ \\ 0, Otherwise \end{matrix}; & 〈 Mathematical .Math. .Math. Equation .Math. .Math. 1 〉 \end{matrix}$

[0056] Wherein, F.sub.i is a second horizontal projection image 4; p.sub.i is a horizontal cumulative value of the corresponding first horizontal projection image 3; μ is an average of the horizontal cumulative values of the first horizontal projection image 3; and σ is a standard deviation.

[0057] The cumulative threshold is defined by Mathematical Equation 2 as follows:

[00005] $\begin{matrix} T = \max (\frac{F_{i}}{ρ}); & 〈 Mathematical .Math. .Math. Equation .Math. .Math. 2 〉 \end{matrix}$

[0058] wherein, ρ is a magnification constant;

[0059] Define a range of the second horizontal projection image 4 having a horizontal cumulative value greater than the cumulative threshold as the search range 5.

[0060] In FIG. 8, the search range 5 is created in the binary image 2.

[0061] In an embodiment, even if the binary image 2 is clear, and the background and the court line 1 can be distinguished, or there is the image of the court line 1 only, or the corner of the court line 1 can be searched directly from the binary image 2, it is preferable to use horizontal projection for processing, since it is always difficult to distinguish the background and the court line 1.

[0062] In another embodiment, the horizontal projection of the first horizontal projection image 3 is performed at the middle third of the binary image 2, since the court line 1 just occupies the middle third of the court image in most court images including audience and advertising, so as to expedite the computation. However, the present invention is not limited by such arrangement only.

[0063] S003: Search a corner of the court line 1 from the search range 5 of the binary image 2 after the search range 5 is created, and use the corners to surround and define a court line range.

[0064] In a preferred embodiment, the plurality of court lines 1 is arranged into a rectangular shape. Due to a possible deviation of angle of a camera occurred while capturing an image, the court line 1 may be distorted into a trapezium or a quadrilateral. In this embodiment, a binary search image 6 and its search coordinates are created, and the search image 6 is divided into an upper left blank 61, an upper right blank 62, a lower left blank 63 and a lower right blank 64 by the search coordinates. For example, the search image 6 of this embodiment is “ custom-character ” shaped.

[0065] In the binary image 2, the white pixel is represented by “1” and the black pixel is represented by “0”, and the white pixel above the pattern (which is the search image 6) is represented by “1” and the black pixel below the pattern is represented by “−1”. The search image 6 is slidably searched in the binary image 2, and a convolution is performed with the binary image 2 as shown in FIGS. 9a to 9d to show the search images 6 of the court line 1 at different situations and the pixel values of the upper left blank 61, the upper right blank 62, the lower left blank 63 and the lower right blank 64. Mathematical Equation 3 is used for the computation as follows:

O(i,j)=Σ.sub.s=−4.sup.4Σ.sub.t=−4.sup.4m(s,t)×p(i+s,j+t);

(i*,j*)=argmax O(i,j); <Mathematical Equation 3>

Wherein, O(i,j) is a corner; m(s,t) is a search image 6; and p(i,j) is a binary image 2.

[0066] Therefore, the maximum obtained after the convolution is defined as the corner, and the corners are used to define a court line range, so as to eliminate the noise outside the court line range. Now, the image within the court line range is just an image including the court line 1.

[0067] S004: Search for a point of intersection 9 of the court line in order to reconstruct the court line 1. Since the photographed court line 1 of the original court image may be covered by dust or the court line 1 in the court image is blocked due to the light factor varied with time in the process of taking the photos. Therefore, a closing process of the image within the court line range as shown in the Steps S201 to S206 of FIG. 10 is performed in this embodiment. In other words, the image within the court line range is dilated to fill the gap of the segment or the broken line of the court line 1, and then an erosion process is performed as shown in FIG. 11. After the closing process, a thinning process of the obtained image is performed as shown in FIG. 12, so that the width of each line is converted into a unit pixel, and then a Hough transform is performed to detect a linear segment 7 of the image obtained after the thinning process, and the linear segment 7 is one intersecting with most points of a straight line on a parametric plane.

[0068] S005: Due to the depression of the net and the fisheye effect of the camera, the image of the court line 1 is distorted, so that segments with repetitions, superimpositions, or noises are produced after the Hough transform takes place, and a filtering process is required. Therefore, the present invention defines at least one first cluster and at least one second cluster according to the characteristics of the linear segment 7 by K-means clustering, and the linear segment 7 is classified into a plurality of groups according to the first cluster and the second cluster, so that the tennis court line just includes vertical lines and horizontal lines only, and the first cluster is used for classifying the horizontal segment in the linear segment 7 according to the slope and the position of the Y-axis coordinate, and the second cluster is used to classify the vertical segment in the linear segment 7 according to the slope and the intercept.

[0069] In the linear segment 7, the court line 1 of the tennis court includes six transverse lines and five vertical lines, so that six first clusters and five second clusters are defined, and the horizontal segments and the vertical segments marked in FIGS. 14 and 15 are classified into similar groups respectively. Since the vertical segments cannot be computed, therefore the vertical segments are created directly by the vertical lines according to the position of the X-axis coordinate.

[0070] S006: Take an average of each group as a standard court line 8, and create linear equations of the standard court line 8 as shown in FIGS. 16, 18 and 19, so as to find the position of the point of intersection 9 of the standard court line 8.

[0071] S007: With reference to FIGS. 17, 20 and 21, the court line 1 of this embodiment is a tennis court line including a net, so that 30 points of intersection 9 can be found. After the points of intersection are found, the court line 1 can be reconstructed according to the points of intersection 9 and the actual position of the court line 1 by a line function.

[0072] However, errors may occur in some situations while reconstructing the court line 1, so that an error filtering step is required. If the court image is a dynamic continuous image, and a reconstruction error occurs, the distance value of the points of intersection 9 compared with the distance value created by the adjacent previous frame will be computed by Mathematical Equation 4 as follows:

|P.sub.i(k+1)−P.sub.i(k)|<T.sub.1,i=1,2, . . . ,30 <Mathematical Equation 4>

[0073] Wherein, T.sub.1 is a first constant value.

[0074] If the point of intersection 9 P.sub.i fits the Mathematical Equation 4, then a computing value is accumulated, and if the computing value is greater than the second constant value, then an error threshold T.sub.e will be computed. In this embodiment, the second constant value is set to 2 as shown in Mathematical Equation 5:

[00006] $\begin{matrix} T_{e} = \frac{α}{30} [{.Math.}_{i = 1}^{30} .Math. .Math. .Math. P_{i} (k + 1) - P_{i} (k) .Math.] & 〈 Mathematical .Math. .Math. Equation .Math. .Math. 5 〉 \end{matrix}$

[0075] Wherein, α is an error magnification constant;

[0076] If the distance value of the points of intersection 9 at the positions of any previous and current frames are greater than the error threshold (or matches Mathematical Equation 6), then the court line 1 of the current frame with a reconstruction error is replaced by the court line 1 reconstructed by the previous frame.

|P.sub.i(k+1)−P.sub.i(k)|>T.sub.e <Mathematical Equation 6>

[0077] In summation, this embodiment as shown in FIGS. 22 to 29 tests different tennis courts at different sites for the tennis match video with a size of 720*480 and a length of 10 seconds, and the backgrounds including the camera position, angle, zooming, background color, advertising, and scoreboard are different, and results show that a complete court line 1 can be captured and reconstructed successfully.

[0078] In this embodiment, the accuracy of reconstructing the court line 1 before performing the error filtering step is approximately 98.4% to 99.7%, and the accuracy of reconstructing the court line 1 after performing the error filtering step reaches 100% (such statistics are obtained from ten thousand videos of open competitions). Obviously, the present invention can locate the point of intersection 9 in the court image accurately to facilitate the reconstruction of the court line 1.

[0079] While this disclosure has been described by means of specific embodiments, numerous modifications and variations could be made thereto by those skilled in the art without departing from the scope and spirit of this disclosure set forth in the claims.

METHOD OF CAPTURING AND RECONSTRUCTING COURT LINES

Inventors

Cpc classification

Classification Explorer

G06T2210/21

PHYSICS

Classification Explorer

G06T11/005

PHYSICS

Classification Explorer

G06T7/13

PHYSICS

Classification Explorer

G06T2207/20061

PHYSICS

Classification Explorer

G06F18/23

PHYSICS

Classification Explorer

G06V10/48

PHYSICS

Classification Explorer

G06V20/42

PHYSICS

Classification Explorer

G06V10/507

PHYSICS

Classification Explorer

G06T5/30

PHYSICS

Classification Explorer

G06T2207/30228

PHYSICS

International classification

Classification Explorer

G06T11/00

PHYSICS

Classification Explorer

G06T5/30

PHYSICS

Classification Explorer

G06K9/46

PHYSICS

Classification Explorer

G06K9/62

PHYSICS

Abstract

Claims

Description