基于深度时空特征模型的非接触式心率估计方法

A Non-Contact Heart Rate Estimation Method Based on a Deep Spatiotemporal Feature Model

为了提升心率计算在不同光照和运动条件下的准确率,提出了一种基于深度时空特征模型的创新方法——非接触式心率计算方法:对输入的人脸视频分别利用卷积神经网络(CNN)提取空间特征和利用长短期记忆网络(LSTM)提取时序特征,将这两种特征融合之后,模型对微弱的血液容积脉冲(BVP)信号展现出更强的表征能力.在自采的ZJXU-Phys数据集上进行验证,新方法在各种条件下均表现出色,其有效性和实用性得到了验证.

In order to improve the accuracy of heart rate calculation under different lighting and exercise conditions,we propose an innovative non-contact heart rate calculation method based on a deep spatiotemporal feature model.We use convolutional neural networks(CNN)to extract spatial features from the input face video,and long short-term memory networks(LSTM)to extract temporal features.By fusing these two features,our model exhibits a stronger representation ability for weak blood volume pulse(BVP)signals.To verify the effectiveness of this method,we conduct experiments on the self-collected ZJXU-Phys dataset,and the experimental results show that the proposed method performs well under various conditions,verifying its effectiveness and practicality.