CN1716281A - Visual quick identifying method for football robot - Google Patents

Abstract

The invention relates to a fast visual identification method for a football robot. It includes installing the camera above the field so that its shooting range completely covers the field; installing the video capture card on the workstation, its input is connected to the output of the camera, the video capture card outputs digital image data, and the digital image is processed by the computer to achieve certainty. The position of the goal and the sideline provides the position, direction, speed and acceleration of the players on the field in real time, as well as the position data of the ball. The program is realized by VC++MFC; the steps are: (1) The feature extraction of the target and the setting of the correction parameters; ( 2) Image preprocessing; (3) Target recognition; (4) Correction processing. The recognition method of the invention has the characteristics of fast search speed, high recognition precision, strong adaptability to the environment and the like.

Description

A Fast Vision Recognition Method for Soccer Robots

technical field

The invention is aimed at digital image processing and recognition when a robot plays a football game. Involves real-time software programming, digital signal processing, computer science and engineering optics and other fields.

Background technique

Robot soccer was formally proposed in 1992 by Professor Alan Mackworth of the University of British Columbia in Canada. This technology and entertainment project emerged in the 1990s and became popular in developed regions such as Europe, America and Japan.

The most important purpose of the Robot Soccer Game is to test the cutting-edge research of artificial intelligence, especially the latest results of multi-agent system research, which can be transformed and applied to other industrial or civilian robots. It can be said that the football robot can be used as a standard platform for researching all robots and artificial intelligence. In addition, football itself is a very entertaining sport, so robot football is an intelligent robot integrating technology and entertainment. It is a cutting-edge scientific research competition and high-tech confrontation based on sports competitions. It is a vivid display of high-tech progress. A window and a new way to promote the practicality and industrialization of scientific and technological achievements.

The game rules of the robot soccer game are similar to the regular football game of human beings. Robots are not controlled by humans during the game and play completely autonomously. In order to achieve this goal, the robot must be composed of the following five parts, namely vision subsystem, decision-making subsystem, wireless communication subsystem, motion control subsystem and mechanical subsystem. The vision and decision-making subsystems are the eyes and brain of the robot, the wireless communication subsystem is the ears and mouth of the robot, the motion control subsystem is equivalent to the human nervous system, and the mechanical subsystem is the hands and feet of the robot.

Among them, the visual subsystem must provide enough information on the field to the decision-making system so that the robot can make accurate judgments on the situation on the field and adopt appropriate strategies. Since the football robot is a highly dynamic system, the speed and accuracy of the visual system processing have a crucial impact on the entire system. There are various drawbacks in the past image recognition methods. It is mainly reflected in the slow search speed, low recognition accuracy, poor adaptability to different environments, and once the light source is changed, target loss and misidentification will occur. This directly leads to deviations in the robot's running position, and even chaos. This is mainly caused by the following reasons:

1. At present, the color model used by most image recognition technologies is the red, green and blue RGB color space. With different light intensities, the same color will also change after being captured by the camera and image card. The RGB model is easily affected by the light source. The system cannot adapt to changes caused by the environment.

2. The wide-angle lens of the camera will produce barrel distortion, and the height difference between the robot and the ball will cause the position deviation of the robot, making the identified robot position inaccurate and affecting the accuracy of image recognition.

3. It will consume a lot of time to recognize all the regions captured by the camera (that is, global recognition). In order to solve this problem, the current image search method is centered on the target position identified in the previous frame, and searches outward in a certain order. However, if the previous frame is not recognized or misidentified, the target will search for the target with a wrong path for a long time, which will affect the recognition speed.

Contents of the invention

The purpose of the present invention is to provide a football robot vision fast recognition method, which has the characteristics of fast search speed, high recognition accuracy and strong adaptability to the environment, and provides enough information on the field for the decision-making system of the football robot.

To achieve the above object, the present invention adopts the following technical solutions:

A kind of football robot vision rapid identification method, comprises that camera is installed above the field so that its shooting range completely covers the field; video capture card is installed on the workstation, its input is connected with the output of camera, and video capture card outputs digitized image data, by The computer processes the digitized image to determine the position of the goal and the sideline, and provides the position, direction, speed and acceleration of the players on the field in real time, as well as the position data of the ball; it is characterized in that the program is realized by VC++MFC, and the steps are as follows:

(1) The feature extraction of the target and the setting of the correction parameters:

The extracted features are color features, and the top cover of each robot is affixed with color codes representing teams and individuals, which are distinguished from the colors of the ball;

The HSV model composed of hue, saturation and brightness is used as the color model for target color feature extraction;

Correction is the spatial geometric transformation of the pixel coordinates of the original image, so that the pixels fall in the correct position, and it is realized in the form of a nonlinear equation. The correction parameters are set on the spot according to the actual site and the position of the camera.

(2) Image preprocessing

Mean filtering of the digitized image to remove mutation interference points.

(3) Target recognition

It is assumed that there are 4 color scales of different colors on each robot, and the position and direction of the robot can be determined through their interrelationships.

(4) Correction processing

The recognized target position is corrected for geometric distortion and height error.

The feature extraction of the above-mentioned target and the setting of the correction parameters are carried out according to the following specific steps:

(1) Collect the image into the internal memory through the image processing card and display it on the screen;

(2) Adjust the contrast, brightness, hue, and color saturation of the image through the library function of the image processing card to change the image acquisition effect until the color scale of the robot and the color of the ball and the field in the image are the largest;

(3) Collect the color information of the ball and the color code in sequence: first, click on the center of the collected object, and the program will display the color of the pixel and the surrounding 7*7 pixels separately; the user will match the actual color of the point After all are selected, the program can get the threshold information of the target HSV; if the user is not satisfied with the value, the HSV threshold range can be changed directly;

(4) Use the mouse to select eight points on the sideline of the field displayed on the computer screen. The program uses the figure surrounded by these eight points to correct a rectangle as a standard, and uses the following correction methods to set the correction coefficient:

(a) Barrel distortion correction:

X'＝K ₁₁ ×(1+K ₁₂ (X ² +Y ² ))X

Y'＝K ₂₁ ×(1+K ₂₂ (X ² +Y ² ))Y

In the formula, K ₁₁ and K ₂₁ are image scale coefficients, K ₁₂ K ₂₂ are distortion correction coefficients;

(b) Camera tilt correction:

X'＝K ₁ ×X/320×Y+X

Y'＝K ₂ ×Y/240×X+Y

In the formula, K ₁ and K ₂ are the tilt correction coefficients in the X and Y directions;

(c) Camera rotation angle correction:

X'＝(X ² +Y ² ) ^1/2 ×cos(arctan(X/Y)+λ)

Y'＝(X ² +Y ² ) ^1/2 ×sin(arctan(X/Y)+λ)

Where λ is the rotation angle.

The above X, Y are the coordinates on the image captured by the camera, and X', Y' are the coordinates after geometric correction; constantly change the correction coefficient until the correction is successful, so as to determine all the correction coefficients; among them, the barrel distortion correction success standard The four sides of the corrected field are all straight lines, the success criterion of the camera tilt parameter setting is that the corrected field is a rectangle, and the successful judgment standard of the camera rotation angle parameter setting is that the corrected four sides of the field are parallel to the image frame.

The above-mentioned target identification is carried out according to the following specific steps:

(1) First, take the ball and the robot team color mark as the target in sequence, and start from the recognized position of the target in the last frame, and search outward in a rectangular spiral manner for pixels with the same color as the target; the rectangular spiral search algorithm is Perform a rectangular spiral search layer by layer around the search center point, and use a circular method to search for points on the four sides to complete a search;

(2) After searching for the team logo of the robot, search for the team member color code indicating the robot number and direction within a certain range centered on the team logo, and judge the specific team member number and direction according to the combination of the identified team member color code;

(3) If it is found that the identified target position and direction do not match the normal logic, it is considered that the target has been misidentified, and misidentification processing is required;

Misidentification is divided into the following three situations:

(a) When a pixel point group exceeding the actual color scale size and meeting the threshold is found, it is judged as misidentification;

(b) If the arrangement of the color code found is not consistent with the actual color code arrangement, that is, no player has the same color code arrangement, it can be considered that the number of the searched player is incorrect and it is determined to be a misidentification;

(c) If the searched robot number is not the current search number, it will be considered as misidentification.

(4) Once the ball or team logo cannot be found within a certain range, a global search algorithm will be used for the target in the next frame, which will block the entire field and search in priority to lose the recognition of the previous frame Location is the highest priority and is searched from near to far.

Compared with the prior art, the present invention has the following obvious outstanding substantive features and significant advantages: the visual recognition method of the present invention is realized in the form of software, and the program is realized by VC++MFC to reach and determine the position of the goal and the sideline, Real-time data such as the position, direction, speed and acceleration of the players on the field and the position of the ball are provided to the decision-making system of the soccer robot. It has the characteristics of fast search speed, high recognition accuracy and strong adaptability to the environment.

Description of drawings

Figure 1 is a workflow block diagram of visual recognition.

Figure 2 is a photogram of the color scale and the ball of the robot.

Figure 3 is a target color threshold sampling screen diagram.

Figure 4 is a block diagram of the calibration parameter tuning process.

Figure 5 is a block diagram of the overall flow of target recognition.

FIG. 6 is a schematic diagram of a rectangular spiral search path with the center point obtained in the previous frame as the seed point.

Fig. 7 is a block diagram of the flow of misrecognition processing.

Fig. 8 is a block diagram of the geometric correction process.

FIG. 9 is a graph showing a comparison before and after correction.

Detailed ways

A preferred example of the present invention is described in detail as follows in conjunction with accompanying drawing:

The football robot vision fast recognition method is implemented in the form of software, and it must be based on a series of hardware foundations. Its basic hardware equipment is a camera and a video capture card. The camera is installed above the field, and the shooting range must completely cover the field. The video capture card is installed on the workstation (it can be a PC), its input is connected to the output of the camera, and the output of the video capture card is digital image data. The present invention is to process and identify these data.

The program is realized by VC++MFC to determine the position of the goal and the sideline, and provide real-time data on the position, direction, speed and acceleration of the players on the field and the position of the ball.

The workflow of the visual recognition method is shown in Figure 1.

1. Feature extraction of target and tuning of correction parameters

To track the target, it is necessary to obtain the features needed to distinguish the target from the non-target to extract the target from the background. The features extracted by the present invention are color features. The top cover of each robot is labeled with team and individual color codes, which are differentiated from the color of the ball. (as shown in Figure 2)

The color model used in the target color feature extraction is the hue·saturation·value (HSV) model. The selection of the color model has a great influence on the correct identification of colors, and HSV is a relatively good color model. The HSV model consists of hue h, saturation s, and brightness v, which is close to the human eye's perception of color. Among them, the hue attribute can reflect the color type more accurately, and is less sensitive to changes in external lighting conditions. For objects with the same color attribute, h has a relatively stable and narrow range of numerical variation, which is used as the main judgment condition. Saturation s is used as an auxiliary judgment condition. The conversion formula from RGB to HSV is as follows:

v = max(r, g, b);

s=1-min(r,g,b)/v;

$\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; =</span>{\mathrm{<span\; class="notranslate">\; cos</span>}}^{<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>\frac{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; 2</span>}\sqrt{{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; g</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; )</span>}}<span\; class="notranslate">\; )</span>$

$\mathrm{<span\; class="notranslate">\; h</span>}<span\; class="notranslate">\; =</span>\left\{\begin{array}{cc}\mathrm{<span\; class="notranslate">\; \&theta;</span>}& <span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; g</span>}\\ \mathrm{<span\; class="notranslate">\; 360</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; \&theta;</span>}& <span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; b</span>}<span\; class="notranslate">\; \&Greater\; Equal;</span>\mathrm{<span\; class="notranslate">\; g</span>}\end{array}\right.$

In addition, due to the non-linearity of the imaging system itself and the different viewing angles of the camera, the image and the actual deviation are caused, and these need to be corrected. Correction is to perform spatial geometric transformation of pixel coordinates on the original image, so that the pixels fall in the correct position. After repeated practice, the present invention realizes this mapping in the form of nonlinear equation. The coefficients of the correction equation must be related to the actual site and the location of the camera, so the correction parameters must be set on site.

The above two parts are completed before the game and require human-computer interaction. It is to tell the robot what color is the ball, what color is the color code of the robot, and how the captured image corresponds to the actual situation. This process is equivalent to training the robot.

The specific steps of target feature extraction and correction parameter tuning are as follows:

(a) The image is collected into the internal memory through the image processing card and displayed on the screen.

(b) Adjust the contrast, brightness, hue and color saturation of the picture through the library function of the image processing card to change the effect of image acquisition until the color scale and the ball on the robot in the image are most distinguished from the field.

(c) Collect the color information of the ball and the color scale sequentially. First, click on the center of the collected object, and the program will display the color of the pixel and the surrounding 7*7 pixels separately (see Figure 3). After the user selects all the points that match the actual color, the program can get the threshold information of the target HSV. If the user is not satisfied with the value, the HSV threshold range can be changed directly.

(d) Use the mouse to select eight points on the displayed edge of the field on the computer screen. The program corrects the figure surrounded by the eight points into a rectangle as a standard, and uses the following correction methods to set the correction coefficient: Barrel distortion correction (caused by the wide-angle lens):

x'=k ₁₁ ×(1+k ₁₂ (x ² +y ² ))x

(k11, k21 are image scale coefficients. K12, k22 are distortion correction coefficients)

y'=k ₂₁ ×(1+k ₂₂ (x ² +y ² ))x

Camera Tilt Correction:

x'=k ₁ ×x/320×y+x

(k1, k2 are the tilt correction coefficients in the x and y directions)

y'=k ₂ ×y/240×x+y

Camera rotation angle correction:

${\mathrm{<span\; class="notranslate">\; x</span>}}^{<span\; class="notranslate">\; ,</span>}<span\; class="notranslate">\; =</span>\sqrt{{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; cos</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; arctan</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; )</span>$

(α is the rotation angle)

${\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{<span\; class="notranslate">\; ,</span>}<span\; class="notranslate">\; =</span>\sqrt{{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; \&times;</span>\mathrm{<span\; class="notranslate">\; sin</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; arctan</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; +</span>\mathrm{<span\; class="notranslate">\; \&alpha;</span>}<span\; class="notranslate">\; )</span>$

The above x, y are the coordinates on the image taken by the camera, and x', y' are the coordinates of the actual situation. All correction coefficients must be determined, and the correction coefficients are constantly changed until the correction is successful. Among them, the success criterion of barrel distortion correction is that the four sides of the field after correction are straight lines. The criterion for judging the success of the camera tilt parameter setting is that the corrected site is a rectangle. The criterion for judging the success of camera rotation angle parameter setting is that the four sides of the field after correction are parallel to the image frame. (See Figure 4 for the tuning flow chart)

2. Image preprocessing

Compared with the actual situation, the image digitized by the video capture card always has a lot of noise. These are determined by the camera and its own precision and external interference. Therefore, the present invention needs to filter the mean value of the digitized image before performing target recognition. The filtering window is 3×3, that is, the HSV value of each pixel is jointly determined by the mean value of the surrounding 3×3 pixels, so that the mutation interference points are removed.

3. Target recognition

The object identified by the invention is a regular geometric figure, and both the ball and the team logo are circular. There are 3 color codes with different colors on each trolley, and the position and direction of the trolley can be determined through their interrelationship. The overall process of target recognition is shown in Figure 5.

The specific steps of target recognition are as follows:

(a) First target the ball and the robot team color code in turn. Starting from the recognized position of the target in the previous frame, search for pixels with the same color as the target in a rectangular spiral manner.

The rectangular spiral search algorithm is to conduct a rectangular spiral search layer by layer around the search center point, and use a circular method to search for points on the four sides to complete a search. This algorithm searches from the center outward, and can conform to the rectangular arrangement of pixels. As shown in the figure, P is the center point obtained in the previous frame as the seed point, and the search sequence is shown by the arrow in Fig. 6 .

(b) After searching for the robot's team logo, search for the team logo indicating the robot's number and direction within a certain range centered on the team logo. Determine the specific team number and direction according to the combination of the identified team members.

(c) If it is found that the identified target position and direction do not match the normal logic, it is considered that the target has been misidentified, and misidentification processing is required.

Misidentification is divided into three cases.

First, when a group of pixel points exceeding the size of the actual color scale and meeting the threshold is found, it is judged as misrecognition.

Second, due to the uniqueness of the arrangement of the identification color codes of our team members, if the arrangement of the searched color codes is inconsistent with the actual color code arrangement, that is, no team member has the same color code arrangement, it can be considered that the searched The player number is incorrect, and it is determined to be a misidentification.

Third, if the searched robot number is not the current search number, it is considered to be misidentified.

For the first case, as long as the pixel point group is not determined as the target.

Misidentification processing is performed on the latter two cases. See Figure 7 for the flow chart.

During the search, record the position information of the recognized object and the number of the player currently searching, and record these search boxes and skip them, and then continue the rectangular spiral search. If a new player number and position information are found, record the position information until this time. The search round ends. If the player to be searched has not been found after the search is over, search for other players. If it is believed that the searched player was found in his own search round, it can be determined that the previously mistakenly searched player is not in the recorded suspected position, and the previous wrongly searched information can be excluded; otherwise, the previous player is indeed in the suspected location. At the position of the searched players, the position information assignment operation is performed on the found players, and the previous suspected information is cleared. After all the team members have searched, check to see if there is any position information that is suspected of being searched by mistake. If so, find the team members who have not been assigned position information in this frame, and perform position assignment operations on them. If there are multiple team members suspected of searching by mistake, you can refer to the position information, direction and movement command information of the previous frame to judge the position of each team member in this frame and perform the assignment operation.

Because once misrecognition occurs, it is impossible to accurately calculate the actual position and direction angle of the player through the identification color mark of the player. The center of the team mark that can only be searched can be used as the position of the player, and the direction of the player needs to be determined by aligning with the previous team mark. The direction of the frame, and the action command issued in the previous frame, and referring to the arrangement of the player's identification color code, calculate an approximate angle as the player's direction in this frame. Although such positioning is not precise enough, at least the position information is given, and the search start position information is provided for the search of the next frame, which is convenient for the search of the next frame.

(d) Once the ball or the team logo cannot be found within a certain range, a global search algorithm will be used for the target in the next frame. This algorithm will block the entire field and search with priority to lose the identification of the previous frame The location is the highest-priority highest-outward search.

4. Correction processing

The calibration flow chart is shown in Figure 8. The recognized target position is geometrically corrected, and the correction equation is described above.

In addition, since the height of the robot cannot be ignored, the identified robot position is its projected position, which deviates from the actual position. Therefore it must be corrected for height, the equation is as follows:

$\mathrm{<span\; class="notranslate">\; r</span>}<span\; class="notranslate">\; =</span>\sqrt{{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}$

r'=r-r/3×h

(h is the height of the robot)

x'=r'×cos(arctan(y/x))

y'=r'×sin(arctan(y/x))

After the above corrections, the site geometry of the image reaches a standard rectangle, and the influence of the height of the robot is eliminated, making the positioning more accurate. The effect before and after correction is shown in Figure 9.

Claims (3)

1. A kind of football robot vision fast recognition method, comprises that video camera is installed on the top of the field so that its shooting range covers the field completely; Video capture card is installed on the workstation, its input connects the output of camera, and video capture card outputs digital image data , the digital image is processed by the computer to determine the position of the goal and the sideline, and the position, direction, speed and acceleration of the players on the field are provided in real time, as well as the position data of the ball; it is characterized in that the program is realized by VC++MFC, and its steps are as follows: (1) The feature extraction of the target and the setting of the correction parameters: The extracted features are color features, and the top cover of each robot is affixed with color codes representing teams and individuals, which are distinguished from the colors of the ball; The HSV model composed of hue, saturation and brightness is used as the color model for target color feature extraction; Correction is the spatial geometric transformation of the pixel coordinates of the original image, so that the pixels fall in the correct position, and it is realized in the form of a nonlinear equation. The correction parameters are set on the spot according to the actual site and the position of the camera. (2) Image preprocessing Mean filtering of the digitized image to remove mutation interference points. (3) Target recognition It is assumed that there are 4 color marks of different colors on the top cover of each robot, and the position and direction of the robot can be determined through their interrelationships. (4) Correction processing The recognized target position is corrected for geometric distortion and height error. 2. the football robot vision quick recognition method according to claim 1, it is characterized in that the feature extraction of described target and the setting of correction parameter are carried out by following specific steps: (1) Collect the image into the internal memory through the image processing card and display it on the screen; (2) Adjust the contrast, brightness, hue and color saturation of the picture through the library function of the image processing card to change the image acquisition effect until the color scale of the robot in the image is distinguished from the color of the ball and the field; (3) Collect the color information of the ball and the color code in sequence: first, click on the center of the collected object, and the program will display the color of the pixel and the surrounding 7*7 pixels separately; the user will match the actual color of the point After all are selected, the program can get the threshold information of the target HSV; if the user is not satisfied with the value, the HSV threshold range can be changed directly; (4) Use the mouse to select eight points on the sideline of the field displayed on the computer screen. The program uses the figure surrounded by these eight points to correct a rectangle as a standard, and uses the following correction methods to set the correction coefficient: (a) Barrel distortion correction: X'＝K ₁₁ ×(1+K ₁₂ (X ² +Y ² ))X Y'＝K ₂₁ ×(1+K ₂₂ (X ² +Y ² ))Y In the formula, K ₁₁ and K ₂₁ are image scale coefficients, K ₁₂ K ₂₂ are distortion correction coefficients; (b) Camera tilt correction: X'＝K ₁ ×X/320×Y+X Y'＝K ₂ ×Y/240×X+Y In the formula, K ₁ and K ₂ are the tilt correction coefficients in the X and Y directions; (c) Camera rotation angle correction: X'＝(X ² +Y ² ) ^1/2 ×cos(arctan(X/Y)+λ) Y'＝(X ² +Y ² ) ^1/2 ×sin(arctan(X/Y)+λ) Where λ is the rotation angle. The above X, Y are the coordinates on the image taken by the camera, and X', Y' are the coordinates after geometric correction; Constantly change the correction coefficient until the correction is successful, so as to determine all the correction coefficients; the standard for the success of barrel distortion correction is that the four sides of the field after correction are straight lines, and the judgment of camera tilt parameters. The criterion for judging the success of parameter setting is that the four sides of the field after correction are parallel to the image frame. 3. the football robot vision fast recognition method according to claim 1, is characterized in that described target recognition is carried out by following specific steps: (1) First, take the ball and the robot team color mark as the target in sequence, and start from the recognized position of the target in the previous frame, and search for pixels with the same color as the target in a rectangular spiral manner; the rectangular spiral search algorithm is Perform a rectangular spiral search layer by layer around the search center point, and use a circular method to search for points on the four sides to complete a search; (2) After searching for the team logo of the robot, search for the team member color code indicating the robot number and direction within a certain range centered on the team logo, and judge the specific team member number and direction according to the combination of the identified team member color code; (3) If it is found that the identified target position and direction do not match the normal logic, it is considered that the target has been misidentified, and misidentification processing is required; Misidentification is divided into the following three situations: (a) When a pixel point group exceeding the actual color scale size and meeting the threshold is found, it is judged as misidentification; (b) If the arrangement of the color code found is not consistent with the actual color code arrangement, that is, no player has the same color code arrangement, it can be considered that the number of the searched player is incorrect and it is determined to be a misidentification; (c) If the searched robot number is not the current search number, it will be considered as misidentification. (4) Once the ball or team logo cannot be found within a certain range, a global search algorithm will be used for the target in the next frame, which will block the entire field and search in priority to lose the recognition of the previous frame Location is the highest priority and is searched from near to far.

Images