摘要
针对运动姿态识别过程中的特征提取效率低的问题,提出一种利用改进贝叶斯优化卷积神经网络(CNN)的运动姿态识别算法。利用关键帧将所采集的人体运动时空数据样本中的不同动作进行划分;采用时空权重自适应插值法对缺失的数据样本中像素值加以补充,再使用卷积神经网络模型提取不同动作姿态的特征并进行矢量化处理;在卷积神经网络模型训练过程中,使用改进的贝叶斯算法对卷积神经网络模型超参数的优化,提高模型的特征提取速度。实验结果表明,所提算法将特征平均提取时间降低了至少20.1%,特征匹配个数提高至98.37%。
Using improved Bayesian optimization convolutional neural network(CNN),a motion posture recognition algorithm is proposed to address the issue of low feature extraction efficiency in the process of motion posture recognition.Keyframes are used to partition different actions in the collected human motion spatiotemporal data samples.The spatio-temporal weight adaptive interpolation method is used to supplement the pixel values in the missing data samples,and then the convolutional neural network model is used to extract the features of different motion postures and vectorize them.In the training process of convolutional neural network model,the improved Bayesian algorithm is used to optimize the super parameters of convolutional neural network model to improve the speed of feature extraction of the model.The experimental results show that the proposed algorithm reduces the average feature extraction time by at least 20.1%,and increases the number of feature matches to 98.37%.
作者
郑永权
董坤
ZHENG Yongquan;DONG Kun(Physical Education Department of the General Education College,Xi’an Jiaotong University City College,Xi’an 710018,China;Department of Physical Education,Northwestern Polytechnical University,Xi’an 710072,China)
出处
《微型电脑应用》
2025年第6期30-33,39,共5页
Microcomputer Applications
基金
陕西省科学技术厅项目(22KRM058)。
关键词
运动姿态识别
卷积神经网络
贝叶斯算法
关键帧
时空权重自适应插值法
motion posture recognition
convolutional neural network
Bayesian algorithm
keyframe
spatio-temporal weight adaptive interpolation method
作者简介
郑永权(1981-),男,硕士,讲师,研究方向为人体运动识别、体育教学信息化;董坤(1978-),男,硕士,副教授,研究方向为体育教育训练学。
