基于3D-CNN和LSTM视觉图像算法的民族传统体育动作识别模型
Ethnic Traditional Sports Action Recognition Model Based on 3D-CNN and LSTM Visual Image Algorithms
徐社远 1朱冰冰1
作者信息
- 1. 合肥经济学院基础课教学部,安徽 合肥 230011
- 折叠
摘要
针对民族传统体育中数据识别中存在的准确率较低、实时性较差等问题,研究将虚拟现实技术与基于三维卷积神经网络的动作捕捉技术结合,并通过长短期记忆神经网络来捕捉动作中的时序信息,提出一种民族传统体育动作识别模型.结果表明,所提出的民族传统体育动作识别模型在最优DroPout比率为0.6时,函数损失收敛在0.021左右,识别精度曲线最终收敛于0.989.与其他姿态识别系统相比,模型识别精度提高超过20%,误差精度低于40 mm.该方法较好地实现了太极拳等民族传统体育项目的动作识别,对民族传统体育的现代化传承与训练起到了促进作用.
Abstract
In response to the problems of low accuracy and poor real-time performance in data recognition in traditional ethnic sports,this study combines virtual reality technology with motion capture technology based on 3D convolutional neural networks,and uses long short-term memory neural networks to capture temporal information in movements,proposing a traditional ethnic sports motion recognition model.The results indicate that the proposed traditional ethnic sports action recognition model converges to a function loss of around 0.021 and a recognition accuracy curve of 0.989 when the optimal DroPaout ratio is 0.6.Compared with other posture recognition systems,the model recognition accuracy has improved by more than 20%,with an error accuracy of less than 40mm.This method has achieved good recognition of movements in traditional ethnic sports such as Tai Chi,and has played a promoting role in the modern inheritance and training of traditional ethnic sports
关键词
民族传统体育/动作识别/模型/3D-CNN/LSTM/视觉图像算法/虚拟现实技术Key words
ethnic traditional sports/action recognition/modeling/3D-CNN/LSTM/visual image algorithms/virtual reality technology引用本文复制引用
出版年
2024
NETL
NSTL
万方数据