CN105664462A - Auxiliary training system based on human body posture estimation algorithm - Google Patents

Abstract

The invention discloses an auxiliary training system based on a human body posture estimation algorithm. First, the background modeling method based on the ViBe model detects and extracts the binary contour map of the human body from the monocular video, and then uses the Canny edge detection algorithm to obtain the contour edge map of the image, and then passes through the horizontal line scanning, human body length ratio constraints and other images The processing method is to obtain the coordinates of the 15 main joint points of the human body in the image. On the basis of the above, build an auxiliary training system, use 5 joint angles composed of 15 joint points as the training index, and select the Euclidean distance as the posture similarity The measurement takes two auxiliary indicators of joint angle trajectory and pose similarity as the system output. The present invention realizes the analysis and comparison of athlete's posture by performing quantitative analysis on sports characteristics, scientifically improves the level and performance of athletes, and makes physical training get rid of the state of purely relying on experience.

Description

Auxiliary Training System Based on Human Pose Estimation Algorithm

Technical field:

The invention relates to an auxiliary training system based on a human body posture estimation algorithm, which belongs to the application of computer vision technology in the field of human motion analysis and sports training.

Background technique:

In traditional sports training, training methods based on naked eye observation are usually used, but with the development of computer vision, people begin to use cameras to capture and analyze athletes' movements. And with the application of computer technology in the field of sports training, a number of companies dedicated to the research of golf auxiliary training systems have emerged at home and abroad. The methods currently used mainly include: image method, map analysis method, and portable sensor method. The existing achievements include the motion software MotionCoach developed by Canadian sports software development company MediaVention, the DartGoffer developed by Swiss sports analysis software development company DartFish, and the GolfSense system launched by Zepp Technology Co., Ltd. However, the degree of digitization of these systems is not enough, and it still needs to consume a lot of manpower and physics to complete, and the practicability is not strong.

The research on golf auxiliary training system is developing towards intelligent and scientific direction, more and more video processing technology and image processing technology are applied in golf swing analysis. Although there are more and more related auxiliary training systems, there are still few intelligent golf swing analysis systems.

Invention content:

In view of this, the present invention provides an auxiliary training system based on the human body posture estimation algorithm, which can be applied in golf auxiliary training and has the advantage of high intelligence. Sports characteristics, realize the analysis and comparison of athletes' posture, scientifically improve the level and performance of athletes, and make sports training get rid of the state of purely relying on experience.

In order to achieve the above object, the present invention proposes an auxiliary training system based on a human body posture estimation algorithm, realizing the transition from an experience-based motion training method to a computer vision-based human motion analysis method; the specific implementation process includes the following steps:

Step (1) target detection; The background modeling method based on ViBe model detects and extracts the binary contour figure of human body in the video;

Step (2) contour edge feature extraction; use the Canny edge detection algorithm to obtain the contour edge map of the image;

Step (3) Estimation of human body posture; through image processing methods such as horizontal line scanning and human body length ratio constraints, to obtain 15 joint point coordinates of the human body model in the contour edge image;

Step (4) Build an auxiliary training system; the system takes two auxiliary indicators of joint angle trajectory and posture similarity as output.

Preferably, in the above-mentioned auxiliary training system based on the human body posture estimation algorithm, based on the posture estimation method based on edge contour features plus image processing, the Canny edge detection algorithm is first used to obtain the contour edge map of the image, and then the horizontal line scanning method is used to extract the head and feet. Joint points: extract knee, hip joint points, neck, and chest joint points based on human body length ratio constraints; obtain hand, elbow, and shoulder joint points based on vertical scanning methods.

Preferably, in the above-mentioned auxiliary training system based on the human body posture estimation algorithm, 5 joint angles composed of 15 joint points are used as the training index, and the Euclidean distance is selected as the similarity measure of the posture, and the joint angle trajectory and the posture similarity are two Auxiliary indicators are taken as the system output; by comparing the joint angle trajectory of the trainer and the coach and according to the similarity of posture, it can be analyzed and guided intuitively.

The auxiliary training system based on the human body posture estimation algorithm in the present invention can be applied in golf auxiliary training, and has the advantage of a high degree of intelligence. It guesses the athlete's action from the image sequence, and realizes the athlete's posture by quantitatively analyzing the motion characteristics. Scientifically improve the level and performance of athletes, make sports training get rid of the state of purely relying on experience, and realize the transformation from experience-based sports training methods to computer vision-based human motion analysis methods.

Description of drawings:

Fig. 1 is a schematic flow chart of the technical solution of the present invention;

Fig. 2 is the flow chart of background modeling method based on ViBe model;

Fig. 3 is the flowchart of Canny edge detection algorithm;

Fig. 4 is a two-dimensional skeleton model of a human body.

detailed description:

The present invention will be further described below in conjunction with the accompanying drawings.

As shown in Figure 1, the pose estimation method based on the edge features of the human body contour plus image processing can obtain the data of 15 joint points of the human body as the input of the auxiliary training system. In order to avoid the difference caused by the height of the person and the distance from the camera, the present invention uses 5 joint angles made up of 15 joint points as system indicators to carry out auxiliary analysis, and the 5 joint angles are respectively recorded as Angle1 (head, neck, Chest), Angle2 (left shoulder, left elbow, left wrist), Angle3 (right shoulder, right elbow, right wrist), Angle4 (left hip, left knee, left foot), Angle5 (right hip, right knee, right foot). Euclidean distance is chosen as the similarity measure of poses. Finally, the system takes two auxiliary indicators of joint angle trajectory and posture similarity as output.

The flow chart of the background modeling method based on the ViBe model is shown in Figure 2. The ViBe algorithm uses neighborhood pixels to create a background model, and detects the foreground target by comparing the pixel value of the current input image with the background model. The background model is initialized; (2) the foreground target segmentation operation is performed on the subsequent image sequence; (3) the background model is dynamically updated in each subsequent frame of pictures by using the eight-neighborhood update method.

When t=k, the background model of the pixel point (x, y) is The pixel value is f ^k (x,y). Determine whether the pixel value is foreground according to the following formula.

${\mathrm{<span\; class="notranslate">\; f</span>}}^{\mathrm{<span\; class="notranslate">\; k</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>\left\{\begin{array}{cc}{\mathrm{<span\; class="notranslate">\; BK</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; ></span>\mathrm{<span\; class="notranslate">\; T</span>}& \mathrm{<span\; class="notranslate">\; f</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; g</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; d</span>}\\ {\mathrm{<span\; class="notranslate">\; BK</span>}}_{\mathrm{<span\; class="notranslate">\; m</span>}}^{\mathrm{<span\; class="notranslate">\; k</span>}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; x</span>}}^{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; ,</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}^{\mathrm{<span\; class="notranslate">\; r</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; \&le;</span>\mathrm{<span\; class="notranslate">\; T</span>}& \mathrm{<span\; class="notranslate">\; b</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; c</span>}\mathrm{<span\; class="notranslate">\; k</span>}\mathrm{<span\; class="notranslate">\; f</span>}\mathrm{<span\; class="notranslate">\; r</span>}\mathrm{<span\; class="notranslate">\; o</span>}\mathrm{<span\; class="notranslate">\; u</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; d</span>}\end{array}\right.<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

The result of the edge extraction of the human body contour by the Canny edge detection operator is shown in Figure 3. The Canny edge detection algorithm is a method of first smoothing and then calculating the derivative. The implementation process is divided into the following steps: (1) smoothing the image with a Gaussian filter; (2) using the first-order difference to calculate the magnitude of the gradient and direction; (3) non-maximum suppression of gradient magnitude; (4) dual threshold detection and connection edges.

Accompanying drawing 4 is a diagram of a two-dimensional skeleton model of a human body, and the coordinates of 15 joint points of the human body model can be obtained through the algorithm of human body posture estimation. The specific implementation process includes the following steps:

(1) Extract head and foot joint points based on horizontal line scanning algorithm.

In the edge contour, draw a horizontal line from the bottom of the image and move it up line by line. Since the human body's feet are at the bottom and are symmetrical, when the horizontal line intersects the edge of the outline for the first time, there should be two intersection points , the intersection point from left to right is marked as the joint point of the left foot and the right foot, and the coordinate of the horizontal line is H1.

Make a horizontal line from the top of the image and move it down line by line. When the horizontal line intersects with the contour edge for the first time, mark this point as the head joint point. At this time, the coordinate of the horizontal line is H2, and the height of the person can be estimated as L=H2-H1.

(2) Extract knee, hip joint points, neck and chest joint points based on length ratio constraints

According to the length ratio constraints of the human body, the length of the feet and knees is 0.286L, and the length of the knees and hips is 0.271L. Make a horizontal line from the foot, and move it up a distance of 0.286L line by line, get four intersection points between the horizontal line and the contour image of the human body edge, mark the midpoints of the two intersection points respectively, and get the left knee and right knee joint points . Then calculate the distance between the foot joint point and the knee joint point, and it must be within the range of the human body scale coefficient, otherwise the result should be corrected. In the same way, the position of the hip joint can also be obtained.

Draw a vertical line at the head joint point, and according to the length ratio constraints of the head, neck, and chest, the coordinates of the neck joint point and chest joint point can be estimated.

(3) Obtain the joint points of the hand, elbow and shoulder based on the vertical scanning algorithm

Make a vertical line from the left part of the image and move it to the right line by line. When the vertical line intersects with the edge of the left contour for the first time, mark this point as the joint point of the left hand. The length between the hand and the elbow is 0.243L, and the length between the elbow and the shoulder The length is 0.2L, draw a vertical line from the joint point of the hand, move it to the right by 0.243L, scan from top to bottom, take the first intersection point and mark it as the left elbow, and move the vertical line from the elbow to the right by 0.2L , scan its first intersection point and record it as the left shoulder. In the same way, the important joint points of the right hand can also be obtained.

Claims (3)

1. the auxiliary training system based on human body attitude algorithm for estimating, it is characterised in that comprise the following steps:

Step (1) target detection; Background modeling method based on ViBe model detects and extracts human body two-value profile diagram in video;

Step (2) contour edge feature extraction; Canny edge detection algorithm is used to obtain the contour edge figure of image;

Step (3) human body attitude is estimated; Through image processing methods such as horizon scan line, human body length ratio constraints, go to obtain 15 body joint point coordinate of anthropometric dummy in contour edge figure;

Step (4) builds auxiliary training system; System is using joint angle track and two auxiliary characteristicss of attitude similarity as output.

2. the auxiliary training system based on human body attitude algorithm for estimating according to claim 1, it is characterized in that: add the Attitude estimation algorithm of image procossing based on edge contour feature, obtain the contour edge figure of image initially with Canny edge detection algorithm, then horizon scan line method extracts head, foot's articulare; Knee, coxa joint point, cervical region, chest articulare is extracted based on the constraint of human body length ratio; Hand, ancon, shoulder joints point is obtained based on vertical scanning method.

3. the auxiliary training system based on human body attitude algorithm for estimating according to claim 1, it is characterized in that: adopt 5 joint angles being made up of 15 articulares as training quota, choose the Euclidean distance measuring similarity as attitude, export as system using joint angle track and two auxiliary characteristicss of attitude similarity; By the joint angle track of comparative training person and coach and according to attitude similarity, thus being analyzed intuitively and guidance.