基于双目视觉的田径运动员犯规动作智能识别方法  

Intelligent Recognition Method of Foul Actions of Track and Field Athletes Based on Binocular Vision

作　　者：蒋清华[1] JIANG Qinghua(Department of Physical Education and Research,Yang'en University,Quanzhou 362014,China)

机构地区：[1]仰恩大学体育教研部,福建泉州362014

出　　处：《长春大学学报》2023年第2期21-26,共6页Journal of Changchun University

基　　金：教育部人文社会科学项目(20YJC890054)。

摘　　要：提出基于双目视觉的田径运动员犯规动作智能识别方法。通过双目相机获取田径运动员动作图像,立体校正双目相机以及实现图像的行对准,获取田径运动员动作深度图像;将提取的前景图像输入双流卷积神经网络模型中,通过批量归一化、非局部特征提取以及A-softmax损失函数,智能识别犯规动作。结果显示:该方法可有效完成图像立体校正,背景扰动影响指标值均低于0.025,精准识别田径运动员犯规动作。An intelligent recognition method based on binocular vision of foul actions of track and field athletes is proposed.The action image of track and field athletes is obtained through binocular camera,the action depth image of track and field athletes is obtained through stereo correction of binocular camera and the line alignment of the image is relized;The extracted foreground image is input into the double stream convolution neural network model,and the foul actions are intelligently recognized through batch normalization,non local feature extraction and Asoftmax loss function.The results show that this method can effectively complete the image stereo correction and the background disturbance influence index values are lower than 0.025,which can accurately identify the foul actions of track and field athletes.

关 键 词：双目视觉 犯规动作 智能识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]