CN114495162A - Volleyball smash action detection method based on video analysis - Google Patents

Abstract

The invention relates to the technical field of intelligent analysis assistance of sports, in particular to a volleyball smash action detection method based on video analysis. The invention can automatically detect the ball-catching action of the athletes in the volleyball event according to the skeleton data of the athletes, and can improve the detection accuracy by utilizing the skeleton joint point data.

Description

Translated fromChinese

一种基于视频分析的排球扣球动作检测方法A volleyball spiking action detection method based on video analysis

技术领域technical field

本发明涉及体育运动智能分析辅助技术领域，具体是一种基于视频分析的排球运动员扣球动作检测方法。The invention relates to the technical field of sports intelligent analysis assistance, in particular to a volleyball player spiking action detection method based on video analysis.

背景技术Background technique

基于计算机视觉的人体目标检测日前已被广泛应用于各类生活场景中，如行人检测、行人轨迹预测、跌倒检测等。在体育运动领域，人体目标检测被用于运动分类、动作检测与运动状态评估，有效提高了识别效率、提高了评估精准度、降低了人力成本。Human target detection based on computer vision has been widely used in various life scenarios, such as pedestrian detection, pedestrian trajectory prediction, fall detection, etc. In the field of sports, human target detection is used for motion classification, motion detection and motion state evaluation, which effectively improves the recognition efficiency, improves the evaluation accuracy, and reduces labor costs.

排球运动作为大球运动的一种，具有排球变化速度快，变化轨迹复杂，场地内运动员密度较大，技术种类多等特点，导致排球运动员的动作变化快，动作种类多。长期以来，用科学技术指导体育训练比较欠缺，给以往通过人眼和经验进行主观判断运动员动作和技术准确性带来巨大的困难。体育运动分析领域迫切需要通过更加快速、标准、准确的更加切实可行的解决方案来更好地实现对排球运动员的姿态识别。因此设计实现基于数字视频检测排球运动员动作检测，提取其中的精彩片段，球队教练可以利用得到的相关数据进行战术研究，将有助于提升竞赛水平的要素付诸于日常球队配合训练中，从而达到最佳的训练效果。Volleyball, as a kind of big ball sports, has the characteristics of fast change speed, complex change trajectory, high density of players in the field, and many types of techniques, which lead to the rapid changes of volleyball players and many types of movements. For a long time, the use of science and technology to guide sports training is relatively lacking, which brings huge difficulties to the subjective judgment of athletes' movements and technical accuracy through human eyes and experience. In the field of sports analysis, there is an urgent need to better realize the gesture recognition of volleyball players through more rapid, standard, accurate and more feasible solutions. Therefore, the design and implementation of the motion detection of volleyball players based on digital video detection, extracting the highlights, the team coach can use the obtained relevant data to conduct tactical research, and put the elements that will help improve the level of competition into the daily team training. in order to achieve the best training effect.

发明专利(发明人:甄新喜，申请号:CN202010262966.8，名称：一种可识别运动员的排球检测系统。)该发明公开了一种基于佩戴式的三轴加速度计的可识别运动员的排球检测系统，通过佩戴在运动员头部、腰部、手腕部和脚腕部以及场地中的CCD光敏传感器检测排球运动员姿态。但基于可穿戴式设备的姿态检测在准确度、检测效率与内存消耗方面远不如基于视觉的方式理想，此外，通过佩戴检测设备在一定程度上也会影响运动员的发挥。发明专利(发明人:周斌，申请号:CN202011306032.6，名称：一种基于改进动态时间规整算法的排球动作识别方法。)该发明公开了一种通过改进的动态时间规整算法对视频中的人体关键点时间序列与标准排球动作人体各关键点相匹配从而判定动作准确度。但是该发明其中骨架关节点数据的提取采用的模型较老，在排球这类变化速度快种类多的运动上效果并不理想，另外，该发明也没有产出对排球辅助训练有帮助的结果。Invention patent (inventor: Zhen Xinxi, application number: CN202010262966.8, name: a volleyball detection system that can identify players.) The invention discloses a volleyball detection system based on a wearable three-axis accelerometer that can identify players , The volleyball player's posture is detected by CCD photosensitive sensors worn on the player's head, waist, wrist and ankle and in the field. However, posture detection based on wearable devices is far less ideal than vision-based methods in terms of accuracy, detection efficiency and memory consumption. In addition, wearing detection devices will also affect the performance of athletes to a certain extent. Invention patent (inventor: Zhou Bin, application number: CN202011306032.6, title: A volleyball action recognition method based on an improved dynamic time warping algorithm.) The invention discloses a method for identifying human bodies in videos through an improved dynamic time warping algorithm The time series of key points is matched with the key points of the standard volleyball action human body to determine the accuracy of the action. However, in this invention, the model used for extracting skeleton joint point data is old, and the effect is not ideal in sports such as volleyball with fast changing speed and variety. In addition, this invention does not produce results that are helpful to volleyball auxiliary training.

综上所述，当前解决方案有一定的局限性，在检测准确度、效率、内存消耗以及后续产出方面不如人意。为此，本发明提出了对排球赛事视频中运动员扣球动作检测的方法。To sum up, the current solutions have certain limitations and are not satisfactory in terms of detection accuracy, efficiency, memory consumption, and subsequent output. Therefore, the present invention proposes a method for detecting a player's spiking action in a volleyball event video.

发明内容SUMMARY OF THE INVENTION

为了解决常规排球赛事中运动员精彩瞬间人工捕捉存在主观判断、效率低下等问题，本发明提供一种基于视频分析的排球运动员扣球动作检测方法。本发明首先使用深度学习的方法直接提取赛事中的运动员骨架关节点数据，然后根据运动员扣球技术规范标准识别处于扣球姿态的排球运动员。In order to solve the problems of subjective judgment and low efficiency in manual capture of players' wonderful moments in conventional volleyball events, the present invention provides a video analysis-based volleyball player spiking action detection method. The invention firstly uses the deep learning method to directly extract the joint point data of the athlete's skeleton in the event, and then recognizes the volleyball player in the spiking posture according to the technical standard of the athlete's spiking.

本发明的技术方案如下：The technical scheme of the present invention is as follows:

一种基于视频分析的排球运动员扣球动作检测方法，包括如下步骤：A volleyball player spiking action detection method based on video analysis, comprising the following steps:

步骤1：利用基于深度学习的人体骨架关节点模型提取排球赛事视频中的排球运动员骨架关节点数据并表示为集合

其中，

表示第k帧中第i个排球运动员的第j个骨架关节点坐标，

表示第k帧中检测到的第i个排球运动员的第j个骨架关节点，其中，

SK为有序集合，其元素分别对应于

的骨架关节点名称；Step 1: Use the deep learning-based human skeleton joint point model to extract the volleyball player skeleton joint point data in the volleyball match video and represent it as a set

in,

represents the coordinate of the jth skeleton joint point of the ith volleyball player in the kth frame,

represents the jth skeleton joint point of the ith volleyball player detected in the kth frame, where,

SK is an ordered set whose elements correspond to

The name of the skeleton joint point;

步骤2：根据式(1)计算第1帧中排球运动员双足的平均高度Y_ave；Step 2: Calculate the average height Y_ave of the feet of the volleyball players in the first frame according to formula (1);

步骤3：针对集合S中的每个骨架关节点数据，根据如下步骤判断排球运动员是否在扣球，具体为：Step 3: For each skeleton joint point data in the set S, determine whether the volleyball player is spiking according to the following steps, specifically:

步骤3.1：根据式(2)计算出排球运动员的双足的平均纵坐标Y_i^k；Step 3.1: calculate the average ordinate Y_i^k of the feet of the volleyball player according to formula (2);

步骤3.2：根据式(3)计算第i个排球运动员的膝关节弯曲角度

根据式(4)计算得肘关节弯曲角度

Step 3.2: Calculate the knee bending angle of the i-th volleyball player according to formula (3).

According to formula (4), the bending angle of elbow joint is calculated

其中，

表示第k帧中第i个排球运动员的第r个关节点与第s个关节点的距离，

和

根据式(5)计算；in,

represents the distance between the r-th joint point and the s-th joint point of the i-th volleyball player in the k-th frame,

and

Calculate according to formula (5);

步骤3.3：根据式(6)计算排球运动员肘部与肩部的高度差，分别记为

和

Step 3.3: Calculate the height difference between the elbow and the shoulder of the volleyball player according to formula (6), and record them as

and

步骤3.4：遍历S中每个排球运动员的骨架关节点数据，判断排球运动员是否在扣球，具体为：对任意第i个排球运动员，若满足条件：Step 3.4: Traverse the skeleton joint point data of each volleyball player in S to determine whether the volleyball player is spiking, specifically: for any ith volleyball player, if the conditions are met:

则判断该运动员正在扣球，其中，α₀为排球运动员扣球膝关节弯曲阈值，β₀为排球运动员扣球手肘弯曲阈值，H₀表示排球运动员扣球需要达到的高度。

Then it is judged that the player is spiking, wherein α₀ is the knee bending threshold of the volleyball player's spiking, β₀ is the elbow bending threshold of the volleyball player's spiking, and H₀ is the height that the volleyball player needs to reach.

本发明的优点为：本发明能够根据运动员骨架数据自动检测出排球赛事中运动员扣球动作，利用骨架关节点数据能够提高检测的准确性，并且能够与其他相似动作例如拦网与传球进行区分。The advantages of the present invention are: the present invention can automatically detect the player's spiking action in a volleyball event according to the player's skeleton data, and the detection accuracy can be improved by using the skeleton joint point data, and can be distinguished from other similar actions such as blocking and passing.

附图说明Description of drawings

图1运动员骨架关节点对应图。Figure 1. Correspondence map of joint points of athlete skeleton.

具体实施方式Detailed ways

下面结合附图和实施例，对本发明进行进一步的说明。应当理解，此处所描述的具体实施例仅用于解释本发明，并不用于限定本发明。The present invention will be further described below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

本发明首先使用深度学习的方法直接提取赛事中的运动员骨架关节点数据，然后根据运动员扣球技术规范标准识别处于扣球姿态的排球运动员。首先收集大量相关排球赛事视频，并提取视频中排球运动员的姿态数据，用于深度网络模型训练，然后使用训练好的深度网络模型进行骨架关节点数据提取，对捕捉到的排球运动员扣球动作进行识别。The invention firstly uses the deep learning method to directly extract the joint point data of the athlete's skeleton in the event, and then recognizes the volleyball player in the spiking posture according to the technical standard of the athlete's spiking. First, collect a large number of relevant volleyball game videos, and extract the pose data of the volleyball players in the videos for training the deep network model. identify.

一种基于视频分析的排球运动员扣球动作检测方法，包括如下步骤：A volleyball player spiking action detection method based on video analysis, comprising the following steps:

步骤1：利用基于深度学习的人体骨架关节点模型提取排球赛事视频中的排球运动员骨架关节点数据并表示为集合

其中，

表示第k帧中第i个排球运动员的第j个骨架关节点坐标，

表示第k帧中检测到的第i个排球运动员的第j个骨架关节点，其中，

SK为有序集合，其元素分别对应于

的骨架关节点名称，如图1所示，

对应鼻1，

对应左眼2，

对应右眼3，

对应左耳4，

对应右耳5，

对应左肩6，

对应右肩7，

对应左肘8，

对应右肘9，

对应左腕10，

对应右腕11，

对应左髋12，

对应右髋13，

对应左膝14，

对应右膝15，

对应左踝16，

对应右踝17；Step 1: Use the deep learning-based human skeleton joint point model to extract the volleyball player skeleton joint point data in the volleyball match video and represent it as a set

in,

represents the coordinate of the jth skeleton joint point of the ith volleyball player in the kth frame,

represents the jth skeleton joint point of the ith volleyball player detected in the kth frame, where,

SK is an ordered set whose elements correspond to

The name of the skeleton joint point, as shown in Figure 1,

Corresponding tonose 1,

Corresponding to theleft eye 2,

Corresponding to theright eye 3,

Corresponding to theleft ear 4,

Corresponding to theright ear 5,

Corresponding to theleft shoulder 6,

Corresponding to theright shoulder 7,

Corresponds to theleft elbow 8,

Corresponds toright elbow 9,

Corresponds toleft wrist 10,

Corresponding to theright wrist 11,

Corresponds to theleft hip 12,

Corresponds toright hip 13,

Corresponding to theleft knee 14,

Corresponding to theright knee 15,

Corresponding to theleft ankle 16,

Corresponds toright ankle 17;

步骤2：根据式(1)计算第1帧中排球运动员双足的平均高度Y_ave；Step 2: Calculate the average height Y_ave of the feet of the volleyball players in the first frame according to formula (1);

步骤3：针对集合S中的每个骨架关节点数据，根据如下步骤判断排球运动员是否在扣球，具体为：Step 3: For each skeleton joint point data in the set S, determine whether the volleyball player is spiking according to the following steps, specifically:

步骤3.1：根据式(2)计算出排球运动员的双足的平均纵坐标Y_i^k；Step 3.1: calculate the average ordinate Y_i^k of the feet of the volleyball player according to formula (2);

步骤3.2：根据式(3)计算第i个排球运动员的膝关节弯曲角度

根据式(4)计算得肘关节弯曲角度

Step 3.2: Calculate the knee bending angle of the i-th volleyball player according to formula (3).

According to formula (4), the bending angle of elbow joint is calculated

其中，

表示第k帧中第i个排球运动员的第r个关节点与第s个关节点的距离，

和

根据式(5)计算；in,

represents the distance between the r-th joint point and the s-th joint point of the i-th volleyball player in the k-th frame,

and

Calculate according to formula (5);

步骤3.3：根据式(6)计算排球运动员肘部与肩部的高度差，分别记为

和

Step 3.3: Calculate the height difference between the elbow and the shoulder of the volleyball player according to formula (6), and record them as

and

步骤3.4：遍历S中每个排球运动员的骨架关节点数据，判断排球运动员是否在扣球，具体为：对任意第i个排球运动员，若满足条件：Step 3.4: Traverse the skeleton joint point data of each volleyball player in S to determine whether the volleyball player is spiking, specifically: for any ith volleyball player, if the conditions are met:

则判断该运动员正在扣球，其中，α₀为排球运动员扣球膝关节弯曲阈值，本实施例中取α₀＝120°，β₀为排球运动员扣球手肘弯曲阈值，本实施例中取β₀＝150°，H₀表示排球运动员扣球需要达到的高度，本实施例中取H₀＝80cm。

Then it is judged that the player is spiking the ball, wherein α₀ is the volleyball player’s spiking knee joint bending threshold, in this embodiment, α₀ =120°, β₀ is the volleyball player’s spiking elbow bending threshold, in this embodiment, take β₀ =150°, H₀ represents the height that the volleyball player needs to reach for the spike, and in this embodiment, H₀ =80cm.

本说明书实施例所述的内容仅仅是对发明构思的实现形式的列举，本发明的保护范围的不应当被视为仅限于实施例所陈述的具体形式，本发明的保护范围也及于本领域技术人员根据本发明构思所能够想到的等同技术手段。The content described in the embodiments of the present specification is only an enumeration of the realization forms of the inventive concept, and the protection scope of the present invention should not be regarded as limited to the specific forms stated in the embodiments, and the protection scope of the present invention also extends to this field. Equivalent technical means that can be conceived by a skilled person according to the inventive concept.

Claims (1)

1. A volleyball smash action detection method based on video analysis is characterized by comprising the following steps:

step 1: skeleton joint point data of volleyball players in volleyball event videos are extracted by utilizing human skeleton joint point models based on deep learning and are expressed as sets

Wherein,

representing the jth skeletal joint coordinate of the ith volleyball player in the kth frame,

represents the jth skeletal joint point of the ith volleyball player detected in the kth frame, wherein,

SK is an ordered set, the elements of which correspond to

The skeleton joint point name of (1);

step 2: calculating the average height Y of the feet of the volleyball player in the 1 st frame according to the formula (1)_ave；

And step 3: and (3) judging whether the volleyball player is catching the volleyball or not according to the data of each skeleton joint point in the set S according to the following steps:

step 3.1: calculating the average ordinate Y of the feet of the volleyball player according to the formula (2)_i^k；

Step 3.2: calculating the knee joint bending angle of the ith volleyball player according to the formula (3)

Calculating the elbow joint bending angle according to the formula (4)

Wherein,

represents the distance between the r-th joint point and the s-th joint point of the ith volleyball player in the k-th frame,

and

calculating according to the formula (5);

step 3.3: calculating the height difference between the elbow and the shoulder of the volleyball player according to the formula (6), and respectively recording the height difference as

And

step 3.4: and traversing the skeleton joint point data of each volleyball player in the S, and judging whether the volleyball player is catching the volleyball, wherein the method specifically comprises the following steps: for any ith volleyball player, if the condition is met:

it is judged that the player is taking a pass, wherein α₀Knee flexion threshold, beta, for volleyball players₀Threshold for volleyball player elbow bend, H₀Indicating the height to which a volleyball player needs to reach for catching a ball。

Images