基于时空图卷积神经网络的网球底线正手击球动作识别方法  

Detection Method of Occluded Athletes in Tennis Game Video Based on Convolution Feature Fusion

作　　者：吴晓军 WU Xiao-jun(Anhui Wenda University of Information Engineering,Hefei,Anhui 231201,China)

机构地区：[1]安徽文达信息工程学院,安徽合肥231201

出　　处：《河北北方学院学报（自然科学版）》2022年第9期14-20,29,共8页Journal of Hebei North University:Natural Science Edition

基　　金：安徽文达信息工程学院校级基金研究项目:“‘健康中国’背景下高校校园足球发展对策研究”(XSK2022A07)。

摘　　要：网球底线正手击球动作识别由于识别准确度低、识别耗时长,导致识别结果无法为教练纠正运动员动作提供有效参考,提出一个基于时空图卷积神经网络的网球底线正手击球动作识别方法。采集图像,建立动作图像采集的输出模板特征匹配模型,采用帧间差分法处理背景,提取图像对应的形状,表征动作变化形态,构建人体有向时空骨架图,建立时空卷积神经网络,参数化处理时空骨架图,将其嵌入到网络,不断迭代,以此完成基于时空图卷积神经网络的网球底线正手击球动作识别。实验结果表明,所提出的识别方法能够准确识别出运动员的正手击球轨迹,在多个动作识别上有效识别出200个正手击球动作,识别时间仅为1.3 min,该方法满足动作识别需求,具备较好的实际应用价值。To improve the detection performance and effect of occluded tennis players,a method of occluded tennis players detection based on convolutional feature fusion was proposed.By extension structure of convolution feature fusion network,image information of shade athletes in the tennis game video was obtained.Then,based on the convolution kernel function feature fusion network,the contour information of shade athlete’s tennis video was extracted.According to the athlete image pixel coordinates,the model of target in image region was constructed with access to the athletes in the tennis match in the video block area.Combined with the pixel gray difference correlation coefficient between two consecutive frames of athlete images,the positioning of occluded athletes in the tennis game video was completed,and the detection results of occluded athletes in the tennis game video were outputted through the training convolution feature fusion network.Experimental results showed that the proposed method not only improves the detection performance,but also has good detection effect.

关 键 词：时空图卷积神经网络 正手 击球动作 识别 网球运动 采集 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]