基于改进DCNN的自由泳技术动作识别  

An Improved DCNN Based on Freestyle Technical Movement Recognition

作　　者：金国利[1] JIN Guoli(School of Physical Education,Quanzhou Normal University,Quanzhou Fujian 362000,China)

机构地区：[1]泉州师范学院体育学院,福建泉州362000

出　　处：《泉州师范学院学报》2024年第5期80-86,共7页Journal of Quanzhou Normal University

摘　　要：为了给自由泳动作的指导和判定提供技术支持,实现对技术动作的精准识别,提出基于向量加权注意力机制改进深度卷积神经网络(DCNN)的识别方法.根据自由泳技术动作规范,设置动作识别标准.从动作视频中提取关键帧图像,引入基于向量加权注意力机制改进DCNN,利用改进算法提取关键帧图像特征,通过特征匹配得出自由泳技术动作的识别结果.实验结果表明:所提方法的动作姿态角识别误差为0.15°,自由泳技术动作识别效果较好,提升了动作类型正确识别率.In order to provide technical support for the guidance and judgment of freestyle movements and achieve accurate recognition of technical movements,a recognition method based on vector weighted attention mechanism and improved deep convolutional neural network(DCNN)is proposed.According to the freestyle swimming technique movement specifications,movement recognition standards were established.Keyframe images were Extracted from action videos,improved DCNN based on vector weighted attention mechanism was introduced,the improved algorithm was used to extract keyframe image features,and recognition results of freestyle technique actions were obtained through feature matching.The experimental results showed that the proposed method had a recognition error of 0.15°for action posture angle,and the freestyle technique had a good effect on action recognition,improving the correct recognition rate of action types.

关 键 词：自由泳 技术动作 深度卷积神经网络 注意力机制 动作识别 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]