CN109030854B - A gait measurement method based on RGB images - Google Patents

Abstract

The invention discloses a pace measurement method based on RGB images, and relates to the field of pace measurement methods. The method includes the following steps: 1: collecting and preprocessing images; The Grabcut algorithm obtains the position of the marker in the image; Step 3: According to the athlete's position, the position of the marker and the set threshold, determine whether the athlete has passed the starting point marker and the ending point marker. The duration of the marker and skip to step 4, if not, repeat this step; Step 4: Calculate the distance between the starting point marker and the ending point marker, and solve the athlete's pace based on the duration of the athlete passing the starting point marker and the ending point marker; The invention realizes accurate positioning of markers, solves the problem of large measurement error caused by lack of distance reference objects in the existing image-based pace measurement methods, and achieves the effect of accurately measuring the pace of athletes.

Description

A gait measurement method based on RGB images

technical field

The invention relates to the field of pace measurement methods, in particular to a pace measurement method based on RGB images.

Background technique

Pedestrian speed measurement is of great significance in real life. The pace measurement in sports track and field events is very important to detect the indicators of athletes. The pace measurement methods can be divided into: 1. Sensor-based pace measurement, 2. Image-based pace measurement. Pace measurement; the sensor-based pace measurement device can measure the user's speed through the speed detection device on the user; there are many ways to measure the pace based on the sensor, Bishop uses two accelerators installed in the calf to calculate the pace, Think of the inverted pendulum pose as a separate stride cycle; Park estimates pace using a handheld device that requires participants to hold a three-axis accelerometer for speed measurements; Gomez uses a wearable visualization sensor device mounted on the user's glasses Test points, use the device to calculate the user's pace; Lu Yongle et al. proposed a multi-motion pattern recognition algorithm based on MEMS inertial sensors, selecting the time domain features of the MEMS acceleration sensor as the pattern recognition feature quantity, and extracting the MEMS angular velocity sensor. The time-domain feature of the sensor is used as a feature quantity for secondary identification to realize pace detection; in the sensor-based method, on the one hand, the tester carries the sensor, which causes it to fail to perform normally, which in turn affects the test result; This makes it impossible to measure the speed in real time. There are also many image-based methods to measure stride speed. For example, Gu et al. used a visual method to track 2D node coordinates with a single RGB camera to estimate stride speed; Zhang Jiajia et al. The Kistler force platform synchronously collected the kinematics and dynamics indicators of fitness race walking, general fitness walking and running; Qin Jianjie and others used the three-camera method and two Sony HXR-NX100 cameras to examine the top 8 athletes during their progress. On the 19th lap, the technical movements at 1km away from the finish line were filmed at a fixed point; Wang Peng et al. used APAS and Dartfish4.5 to analyze the video, used the Japanese human body inertial parameter model, and selected 19 joint points for digital research A compound step of the object, the 3D is smoothed by low-pass digital filtering with a cutoff frequency of 8 Hz, and the competition image is processed with Dartfish software.

Although a large number of research results have been achieved in the measurement of pace, there is no distance reference in the image in the image-based method. The actual distance passed by the person is calculated by the position of the person in the RGB image, and the obtained distance will be affected by the camera's own parameters, image Due to the influence of quality and other factors, the error of the pace measurement result is large. Therefore, there is a need for a high-accuracy image-based gait measurement method.

SUMMARY OF THE INVENTION

The purpose of the present invention is: the present invention provides a pace measurement method based on RGB images, which solves the problem that the lack of distance reference objects in the image in the existing image-based pace measurement method, which is influenced by various factors, leads to large errors in the measurement results The problem.

The technical scheme adopted in the present invention is as follows:

A method for measuring stride speed based on RGB images, comprising the following steps:

Step 1: Collect and preprocess images;

Step 2: Obtain the position of the athlete in the image through HOG feature detection, and obtain the position of the marker in the image through the Grabcut algorithm;

Step 3: Determine whether the athlete has passed the starting point marker and the ending point marker according to the athlete's position, the position of the marker and the set threshold value. No, repeat this step;

Step 4: Calculate the distance between the starting point marker and the ending point marker based on step 2, and calculate the athlete's pace based on the duration of the athlete passing the starting point marker and the ending point marker.

Preferably, the step 1 includes the following steps:

Step 1.1: Determine the vertical distance D between the camera and the marker according to the height H of the marker and the imaging height h of the marker in the camera. The calculation formula is as follows:

Among them, f represents the focal length of the camera;

Step 1.2: Install a fixed camera for image acquisition according to the vertical distance between the camera and the marker;

Step 1.3: Perform scale transformation, grayscale and other operations on the collected image to obtain a preprocessed image.

Preferably, obtaining the position of the athlete in the image through HOG feature detection in the step 2 includes the following steps:

Step S2.1.1: Use the HOG feature to form a sliding window, according to the horizontal (x direction) rightward and vertical (y direction) upward direction, after determining the step size according to the filling size, the feature extraction is performed on the preprocessed image, Calculated as follows:

G(x _i ,y _i )=d _x (x _i ,y _i )+d _y (x _i ,y _i )

d _x (x _i ,y _i )=I(x _i +1,y _i )-I(x _i ,y _i )

d _y (x _i ,y _i )=I(x _i ,y _i +1)-I(x _i ,y _i )

Among them, I(x _i , y _i ) represents the pixel value of the ith point of the input image, G(x _i , y _i ) represents the gradient of the input image, and d _x (x _i , y _i ) represents the horizontal gradient of the input image , dy (x _i , _{y i} ) represents the vertical gradient of the input image;

Step S2.1.2: Input the extracted features into the trained model SVM, and obtain multiple rectangular frames of the area where the athlete may be located in the image;

Step S2.1.3: Obtain the optimal frame in multiple rectangular frames through non-maximum suppression NMS, that is, calculate the intersection ratio IOU of multiple rectangular frames, and select the rectangular frame with the smallest intersection ratio IOU as the optimal frame;

Step S2.1.4: Athlete detection: Calculate the center of the athlete's position through the coordinates of the optimal frame to obtain the athlete's position in the image. The calculation formula is as follows:

Among them, (x _i , y _i ) represents the vertex coordinates of the optimal frame, N represents the number of vertices, x _ci represents the abscissa of the athlete's position, and y _ci represents the ordinate of the athlete's position.

Preferably, in the step 2, obtaining the position of the marker in the image by the Grabcut algorithm includes the following steps:

Step S2.2.1: mark the region of interest, ie ROI, on the preprocessed image;

Step S2.2.2: Use the Grabcut algorithm to segment the ROI, obtain the source node that is the foreground sign, the sink node that is the background sign and the pixels of the foreground image, and create a new map containing all the pixels of the foreground image of the ROI, where the foreground refers to the marker;

Step S2.2.3: After multiple iterations, the pixels connected to the source node are used as the foreground, and the pixels connected to the sink node are used as the background to complete the segmentation of the image foreground and background, and obtain the upper left corner coordinates of the foreground image through the formula, The width and height are calculated as follows:

pick(x, y, w, h)

Among them, x represents the abscissa of the upper left corner of the foreground image, y represents the ordinate of the upper left corner of the foreground image, w represents the width of the foreground image, and h represents the height of the foreground image;

Step S2.2.4: Calculate the abscissa of the marker through the abscissa of the upper left corner of the foreground image and its width. The calculation formula is as follows:

_xbi = x+w/2

Among them, _xbi represents the abscissa of the marker.

Preferably, the step 3 includes the following steps:

Step 3.1: Compare and analyze the relative positions of the athlete and the marker in the image, calculate the difference between the abscissa _xci of the athlete and the abscissa _xbi of the marker, and compare the difference with the set threshold σ , after taking the position corresponding to the first difference less than the threshold value in the forward direction of the athlete as the timing point, determine that the athlete passes the first marker, the starting point marker, and the second marker, the ending point marker, and record the passing of the starting point marker. and the time of the endpoint marker, the calculation formula is as follows:

Among them, t _i represents the time to pass the marker, _ts represents the time to pass the starting point marker, t _e represents the time to pass the end marker, x _ci represents the abscissa of the athlete in the current frame, x _b1 represents the time of the first marker abscissa, x _b2 represents the abscissa of the second marker, σ represents the set threshold;

Step 3.2: Calculate the duration t _f through the start and end markers:

Among them, t _p represents the time required to process each frame of pictures.

Preferably, the step 4 includes the following steps:

Step 4.1: According to the distance D between the camera and the marker, and the camera's field of view angle θ is known, calculate the distance L between the starting point marker and the ending point marker. The calculation formula is as follows:

Step 4.2: Calculate the pace of the athlete according to the distance L between the starting point marker and the ending point marker and the duration _tf passing through the starting point marker and the ending point marker. The calculation formula is as follows:

Among them, λ represents the stride length of the average athlete.

Preferably, the step length is calculated according to the height of the athlete, and the calculation formula is as follows:

SG=λ*0.54+132

Among them, SG represents the height in cm, and λ represents the step length of an ordinary athlete.

Preferably, the calculation formula of the set threshold σ is as follows:

Among them, σ represents the set threshold, W _f represents the pixel value of the image width, W _r represents the width of the camera lens, v represents the running speed, and fps represents the number of frames per second of the video.

To sum up, due to the adoption of the above-mentioned technical solutions, the beneficial effects of the present invention are:

1. the present invention determines the position of the marker in the video/image by the Grabcut algorithm, detects the position of the athlete in the image by the HOG feature, determines the start time and the end time according to the athlete's position, the position of the marker and the threshold value set, that is, determines that the athlete passes through. Marker, obtain the time of passing the marker, and solve the athlete's pace by combining the distance passing through the marker, which solves the problem of the lack of distance reference in the image in the existing image-based pace measurement method, which is affected by various factors and causes measurement error. A big problem, to achieve the effect of accurately positioning the marker and accurately calculating the pace of the athlete;

2. The present invention needs to overcome that the HOG feature algorithm is affected by the height and body shape of the athlete, and the step length is calculated according to the height when calculating the pace to further reduce the measurement error;

3. The present invention reasonably sets the vertical distance between the athlete and the camera and the horizontal distance between the markers, so as to avoid the disadvantage that the marker cannot be completely photographed by the camera because the distance is too far, and the instantaneous speed of the athlete is measured, resulting in a larger error;

4. The measurement object of the present invention is an athlete, and its speed differs greatly from the general measurement object, and the distance traveled by the athlete in each frame image is related to the speed, and the threshold is reasonably set to judge the abscissa difference between the position of the marker and the position of the athlete. With the size of the threshold, it is determined that the athlete passes the marker, so as to calculate the corresponding time and distance, avoiding the disadvantage of directly calculating the distance of the person passing in the image, which is affected by camera factors and causing large measurement errors, and accurately locating the marker is conducive to improving the calculation of pace. accuracy.

Description of drawings

The invention will be described by way of example and with reference to the accompanying drawings, in which:

Fig. 1 is the method flow chart of the present invention;

Fig. 2 is the method flow chart of the present invention;

Fig. 3 is the imaging principle diagram of the present invention;

Fig. 4 is the schematic diagram of camera placement of the present invention;

Fig. 5 is the flow chart of pace calculation of the present invention;

Fig. 6 is athlete detection flow chart of the present invention;

FIG. 7 is a flowchart of marker detection of the present invention.

Detailed ways

All features disclosed in this specification, or all disclosed steps in a method or process, may be combined in any way except mutually exclusive features and/or steps.

The present invention will be described in detail below with reference to FIGS. 1-7 .

The technical problem to be solved by this application: obtain the position of the athlete in the image through HOG feature detection, obtain the position of the marker in the image through the Grabcut algorithm, and judge whether the athlete passes the starting point/ End marker, obtain the duration of passing through the starting marker and the ending marker, calculate the distance between the starting marker and the ending marker, so as to obtain the athlete's pace; Step 2: Obtain the position of the athlete in the image through HOG feature detection, and obtain the position of the marker in the image through the Grabcut algorithm; Step 3: Determine whether the athlete has passed the starting point mark according to the position of the athlete, the position of the marker and the set threshold If it is, then calculate the duration of the athlete passing the starting point marker and the ending point marker and skip to step 4. If not, it means that the timing cannot be started, and step 3 needs to be repeated until the difference is less than the threshold, and record this time. Step 4: Calculate the distance between the starting point marker and the ending point marker based on step 2, and calculate the athlete's pace based on the time the athlete passes through the starting point marker and the ending point marker. Among them, it is determined that the athlete is determined by comparing the abscissa of the position of the athlete and the position of the marker by comparing the abscissa of the starting point marker and the ending point marker to be less than the set threshold, and the corresponding time is recorded after the determination; the distance between the markers needs to be overcome during the measurement process. To ensure the rationality of the vertical distance between the position of the camera and the marker, to avoid the athlete being too small or too large in the image, the distance between the camera and the athlete needs to be set reasonably.

Example 1

Acquisition of data: acquisition of images/videos of athletes and markers through multimedia devices;

Preprocessing: perform scale transformation and grayscale operations on the collected images/videos of athletes and markers, and remove the influence of illumination and noise to complete the preprocessing;

Feature extraction: Use HOG for feature extraction on athlete images/videos; build a model GMM for the RGB three-channel of the marker images/videos for feature extraction, and mark the region of interest (ROI); GMM is used to model the foreground and background. At the next iteration, it learns and creates a new pixel distribution and classifies the foreground and background based on the foreground or background pixel values from the previous iteration.

Classification: Put the extracted athlete image/video features into the trained model SVM to obtain multiple rectangular boxes of the possible area of the athlete, and select the optimal box to calculate the athlete's position in the image; use the Grabcut algorithm to segment the ROI, Obtain the source node that is the foreground sign, the sink node that is the background sign and the pixels of the foreground image, and create a new map containing all the pixels of the ROI foreground image, where the foreground refers to the marker; after several iterations, the accurate segmentation of the image foreground and background is completed, and Calculate the location of the marker in the image;

Time measurement: By comparing the relative positions of the athlete and the marker, judging that the athlete passes the marker and recording the passing time, the first marker passed is the starting point marker, and the second marker is the end marker, and it is judged that the athlete has passed the marker The marker is to compare the difference between the abscissa of the athlete's position and the abscissa of the marker's position and the set threshold value. The calculation formula of the comparison is as follows:

Among them, t _i represents the time to pass the marker, _ts represents the time to pass the starting point marker, t _e represents the time to pass the end marker, x _ci represents the abscissa of the athlete in the current frame, x _b1 represents the time of the first marker abscissa, x _b2 represents the abscissa of the second marker, σ represents the set threshold;

Calculate the duration _tf through the start and end markers:

Among them, t _p represents the time required to process each frame of pictures.

Calculate pace: Calculate pace according to the formula according to the duration of passing the starting point marker and the ending point marker and the distance between the starting point marker and the ending point marker:

Among them, L represents the distance between the starting point marker and the ending point marker, t _f represents the time length of passing the starting point marker and the ending point marker, and λ represents the step length of an ordinary athlete.

Example 2

Collection and preprocessing:

The imaging principle diagram shown in Figure 3:

Step 1.1: Determine the vertical distance D between the camera and the marker according to the height H of the marker and the imaging height h of the marker in the camera. The calculation formula is as follows:

Among them, f represents the focal length of the camera, and its value is 45mm; the height H of the marker is 0.75m; the original image size is 1280×720, and the resolution is 300dpi, which is 300 pixels per inch, so the pixel size is: (300*2.54 )/(1280*720)=0.0083cm/pixel; the marker occupies 85 pixels in the vertical direction of the image, and the imaging height h of the marker can be obtained as 0.7cm, thus the vertical distance between the camera and the marker D= 4.9m;

Step 1.2: Install a fixed camera for image acquisition according to the vertical distance between the camera and the marker;

Step 1.3: Perform scale transformation, grayscale and other operations on the collected image to complete the preprocessing.

Athlete detection:

Step S2.1.1: Use the HOG feature to form a sliding window, according to the horizontal (x direction) rightward and vertical (y direction) upward direction, after determining the step size according to the filling size, the feature extraction is performed on the preprocessed image, Calculated as follows:

G(x _i ,y _i )=d _x (x _i ,y _i )+d _y (x _i ,y _i )

d _x (x _i ,y _i )=I(x _i +1,y _i )-I(x _i ,y _i )

d _y (x _i ,y _i )=I(x _i ,y _i +1)-I(x _i ,y _i )

Among them, I(x _i , y _i ) represents the pixel value of the ith point of the input image, G(x _i , y _i ) represents the gradient of the input image, and d _x (x _i , y _i ) represents the horizontal gradient of the input image , d _y (x _i , y _i ) represents the vertical gradient of the input image; according to a single pixel interval as the step size, the smaller the pixel interval, the wider the image traversal degree, which further improves the accuracy of feature extraction, and the image padding is N, The horizontal and vertical steps of the sliding window are set to N/2, and the image is enlarged by 1.05 times. In this embodiment, the padding is set to 8, and the horizontal and vertical steps are set to 4;

Step S2.1.2: Input the extracted features into the trained model SVM to obtain multiple rectangular boxes in the area where the athlete may be located;

Step S2.1.3: Obtain the optimal frame in multiple rectangular frames through non-maximum suppression (NMS), that is, calculate the intersection ratio IOU of multiple rectangular frames, and select the rectangular frame with the smallest intersection ratio IOU as the optimal frame;

Step S2.1.4: Athlete detection: Calculate the center of the athlete to obtain the position of the athlete in the image through the coordinates of the optimal frame. The calculation formula is as follows:

Among them, (x _i , y _i ) represents the vertex coordinates of the optimal frame, N represents the number of vertices, x _ci represents the abscissa of the athlete's position, and y _ci represents the ordinate of the athlete's position.

Marker detection:

Step S2.2.1: mark the region of interest, ie ROI, on the preprocessed image;

Step S2.2.2: Use the Grabcut algorithm to segment the ROI, obtain the source node that is the foreground sign, the sink node that is the background sign, and the pixels of the foreground image, and create a new map containing all the pixels of the ROI foreground image, where the foreground refers to the marker, Grabcut The algorithm is to establish a Gaussian mixture model GMM for the RGB three channels. The GMM is as follows:

π_k表示N(x|μ_k,Σ_k)的权重；GMM用于对前景和背景建模，在每次迭代时，其学习并创建新的像素分布，并且对前景和背景进行分类，分类的依据是上一次迭代中前景或背景像素值，Grabcut算法通过不断地迭代使前景图像更加精准。Among them, N(x|μ _k ,Σ _k ) represents the kth element of GMM, and K is the mixing coefficient, which satisfies

π _k represents the weight of N(x|μ _k ,Σ _k ); GMM is used to model the foreground and background, at each iteration, it learns and creates a new pixel distribution, and classifies the foreground and background, classification The basis is the foreground or background pixel value in the previous iteration, and the Grabcut algorithm makes the foreground image more accurate through continuous iteration.

Step S2.2.3: After multiple iterations, the pixels connected to the source node are used as the foreground, and the pixels connected to the sink node are used as the background to complete the segmentation of the image foreground and background, and obtain the upper left corner coordinates of the foreground image through the formula, The width and height are calculated as follows:

pick(x, y, w, h)

Among them, x represents the abscissa of the upper left corner of the foreground image, y represents the ordinate of the upper left corner of the foreground image, w represents the width of the foreground image, and h represents the height of the foreground image;

Step S2.2.4: Calculate the abscissa of the marker through the abscissa of the upper left corner of the foreground image and its width. The calculation formula is as follows:

_xbi = x+w/2

Among them, _xbi represents the abscissa of the marker.

Determine how long the athlete passes the start and finish markers:

Step 3.1: Compare and analyze the relative positions of the athlete and the marker in the image, calculate the difference between the abscissa of the athlete and the abscissa of the marker, and compare the difference with the set threshold. If the difference is less than the threshold, the first difference position in the forward direction of the athlete is selected as the timing point. When the difference between the abscissas of the two markers and the athlete is less than the set threshold σ, it is determined that the athlete has passed the first marker. That is, the starting point marker and the second marker, the end point marker, and the time passing through the starting point marker and the end point marker is recorded. The calculation formula is as follows:

Among them, t _i represents the time to pass the marker, _ts represents the time to pass the starting point marker, t _e represents the time to pass the end marker, x _ci represents the abscissa of the athlete in the current frame, x _b1 represents the time of the first marker The abscissa, x _b2 represents the abscissa of the second marker, σ represents the set threshold; the set threshold σ can be obtained according to the formula, and the specific calculation formula is as follows: The set threshold σ calculation formula is as follows:

Among them, σ represents the set threshold, W _f represents the pixel value of the image width, W _r represents the width of the camera lens, v represents the running speed, and fps represents the number of frames per second of the video;

In this embodiment, the parameters for calculating the threshold are all known or given according to the actual situation. This application will calculate the threshold in combination with these parameters to ensure the accuracy of the setting of the threshold; according to the investigation, the fastest speed of the 100-meter race is known. That is, v is 10m/s, the athlete’s walking video is 25fps, the camera is in the 1280*720 scale, the pixel value of the image width W _f is 1280, and the lens width W _r is 10m. Therefore, from the above data, it can be concluded that the athlete is in the forward direction. The maximum pixel value of each frame movement is 51.2, so the set threshold is 50, which can effectively detect whether a person passes through the marker.

Step 3.2: Calculate the duration t _f through the start and end markers:

Among them, t _p represents the time required to process each frame of pictures.

Calculate the pace: as shown in Figure 4:

Step 4.1: Calculate the distance L between the markers according to the distance D between the camera and the marker, and the camera's field of view angle θ is known. The calculation formula is as follows:

The actual measurement can get the camera's field of view angle θ=90°, which can be L=9.8m;

Step 4.2: Calculate the pace of the athlete according to the distance between the starting point marker and the ending point marker obtained in step 4.1 and the duration of the starting point marker and the ending point marker. The calculation formula is as follows:

Among them, L represents the distance between the markers, t _f represents the duration of passing the starting point marker and the ending point marker, and λ represents the step length of an ordinary athlete.

The step length is calculated according to the height of the athlete, and the calculation formula is as follows:

SG=λ*0.54+132

Among them, SG represents the height in cm, and λ represents the step length of an ordinary athlete.

For example, the height is 178cm, the actual step length is 85cm, and the calculated value is 85.18cm. This formula is obtained through experiments by multiple athletes, statistics and analysis, and the accuracy is high.

To sum up, the present invention determines the position of the marker in the video/image through the Grabcut algorithm, detects the position of the athlete in the image through the HOG feature, and determines the start time and end time according to the athlete's position, the position of the marker and the set threshold, that is, the athlete is determined. Through the marker, the duration of passing the marker is obtained, and the athlete's pace is calculated based on the distance passing through the marker, which solves the problem of the lack of distance reference in the image in the existing image-based pace measurement method, which is affected by various factors. The problem of large error has achieved the effect of accurately positioning the marker and accurately calculating the pace of the athlete.

Claims (8)

1. A pace measuring method based on RGB images is characterized in that: the method comprises the following steps:

step 1: collecting and preprocessing an image;

step 2: obtaining the position of the athlete in the image through HOG characteristic detection, and obtaining the position of the marker in the image through Grabcut algorithm;

and step 3: judging whether the athlete passes through the starting point marker and the end point marker according to the position of the athlete, the position of the marker and a set threshold value, if so, passing through a formula

Calculating the time length of the athlete passing through the starting point marker and the end point marker, jumping to the step 4, if not, repeating the step, wherein t_pRepresenting the time required to process each frame of the picture, t_sIndicates the time, t, of passage of the origin marker_eIndicates the time to pass the end-point marker, t_fThe length of time for the start marker and the end marker;

and 4, step 4: based on step 2 by formula

Calculating the distance between the start point marker and the end point marker, and combining the time length of the athlete passing through the start point marker and the end point marker by a formula

And solving the walking speed of the athlete, wherein D is the distance between the camera and the marker, theta is the visual angle of the camera, L is the distance between the starting point marker and the end point marker, and lambda represents the step length of the ordinary athlete.

2. The RGB image-based pace measuring method according to claim 1, wherein: the step 1 comprises the following steps:

step 1.1: according to the height H of the marker and the imaging height H of the marker in the camera, the vertical distance D between the camera and the marker is determined, and the calculation formula is as follows:

wherein f represents the focal length of the camera;

step 1.2: installing and fixing a camera according to the vertical distance between the camera and the marker to acquire an image;

step 1.3: and carrying out scale transformation and graying operation on the acquired image to obtain a preprocessed image.

3. The RGB image-based pace measuring method according to claim 2, wherein: the step 2 of obtaining the position of the athlete in the image through the HOG feature detection comprises the following steps:

step S2.1.1: forming a sliding window by using HOG characteristics, determining the step length according to the filling size and the horizontal (x direction) right direction and the vertical (y direction) upward direction, and then performing characteristic extraction on the preprocessed image, wherein the calculation formula is as follows:

G(x_i，y_i)＝d_x(x_i，y_i)+d_y(x_i，y_i)

d_x(x_i，y_i)＝I(x_i+1，y_i)-I(x_i，y_i)

d_y(x_i，y_i)＝I(x_i，y_i+1)-I(x_i，y_i)

wherein, I (x)_i，y_i) The pixel value, G (x), representing the ith point of the input image_i，y_i) Representing the gradient of the input image, d_x(x_i，y_i) Representing the horizontal gradient of the input image, d_y(x_i，y_i) Representing a vertical gradient of the input image;

step S2.1.2: inputting the extracted features into a trained model SVM to obtain a plurality of rectangular frames of an area in which the athlete is likely to be located in the image;

step S2.1.3: obtaining an optimal frame in the plurality of rectangular frames through the non-maximum inhibition NMS, namely calculating the intersection ratio IOU of the plurality of rectangular frames, and selecting the rectangular frame with the minimum intersection ratio IOU as the optimal frame;

step S2.1.4: detecting athletes: calculating the position of the athlete in the image by calculating the center of the position of the athlete according to the coordinates of the optimal frame, wherein the calculation formula is as follows:

wherein (x)_i，y_i) Representing the coordinates of the vertices of the optimal box, N representing the number of vertices, x_ciAbscissa, y, representing the position of the player_ciThe ordinate representing the player position.

4. The RGB image-based pace measuring method according to claim 3, wherein: the step 2 of obtaining the position of the marker in the image through the Grabcut algorithm comprises the following steps:

step S2.2.1: marking a region of interest (ROI) on the preprocessed image;

step S2.2.2: segmenting the ROI by using a Grabcut algorithm, obtaining source nodes, namely foreground marks, sink nodes, namely background marks, and foreground image pixel points, and establishing a new image containing all ROI foreground image pixel points, wherein the foreground refers to a mark;

step S2.2.3: and (3) taking the pixel point connected with the source node as a foreground through multiple iterations, taking the pixel point connected with the sink node as a background, completing the segmentation of the foreground and the background of the image, and obtaining the coordinates, the width and the height of the upper left corner of the foreground image through a formula, wherein the calculation formula is as follows:

pick(x，y，w，h)

wherein x represents the abscissa of the upper left corner of the foreground image, y represents the ordinate of the upper left corner of the foreground image, w represents the width of the foreground image, and h represents the height of the foreground image;

step S2.2.4: and calculating the abscissa of the marker through the abscissa of the upper left corner of the foreground image and the width of the foreground image, wherein the calculation formula is as follows:

x_bi＝x+w/2

wherein x is_biThe abscissa of the marker is indicated.

5. The RGB image-based pace measuring method according to claim 4, wherein: the step 3 comprises the following steps:

step 3.1: comparing and analyzing the relative position of the athlete and the marker in the image according to the abscissa x of the athlete_ciAnd the abscissa x of the marker_biCalculating the difference between the two, comparing the difference with a set threshold value sigma, taking the position corresponding to the difference of the first advancing direction of the athlete smaller than the threshold value as a timing point, determining that the athlete passes through the first mark, namely a starting point mark, and the second mark, namely an end point mark, and recording the time of passing through the starting point mark and the end point mark, wherein the calculation formula is as follows:

wherein, t_iIndicates the time of passage of the marker, t_sIndicates the time, t, of passage of the origin marker_eIndicates the time of passage of the end-point marker, x_ciRepresenting the abscissa, x, of the athlete at the current frame_b1Denotes the abscissa, x, of the first marker_b2Represents the abscissa of the second marker, and σ represents the set threshold;

step 3.2: calculating the time length t of passing through the start point marker and the end point marker_f：

Wherein, t_pRepresenting the time required to process each frame of the picture.

6. The RGB image-based pace measuring method according to claim 5, wherein: the step 4 comprises the following steps:

step 4.1: according to the distance D between the camera and the marker, knowing the angle of view theta of the camera, calculating the distance L between the starting point marker and the end point marker, wherein the calculation formula is as follows:

step 4.2: according to the distance L of the starting point marker and the end point marker and the time length t passing through the starting point marker and the end point marker_fCalculating the walking speed of the athlete according to the following calculation formula:

where λ represents the step size of the average athlete.

7. The RGB image-based pace measuring method according to claim 6, wherein: the step length is calculated according to the height of the athlete, and the calculation formula is as follows:

SG＝λ*0.54+132

wherein SG represents height in cm, and lambda represents the step length of an ordinary athlete.

8. The RGB image-based pace measuring method according to claim 5, wherein: the set threshold value σ is calculated as follows:

where σ denotes a set threshold value, W_fPixel value, W, representing the width of an image_rRepresenting camera lens extent, v running speed, fps video frames per second.

Images