EEMD降噪的行人活动识别方法

Pedestrian activity recognition method based on EEMD noise reduction

针对微机电传感器加速度计存在随机误差引起的数据质量差,从而导致行人活动识别精度不高的问题,提出了一种基于集合经验模态分解算法的数据降噪行人活动识别方法。利用加速度计收集训练数据和活动识别数据,采用集合经验模态分解算法,将加速度信号分解为多个本征模态分量和一个残余分量,去除高频部分模态分量并对其余分量进行重构,得到降噪信号后再提取时域、频域特征,分别进行模型训练和活动识别。三组行人活动识别实验结果表明,基于集合经验模态分解算法的数据降噪行人活动识别方法对5种日常行人活动识别率较高,平均查全率达到91.8%,平均查准率到达92.3%,其中慢跑查全率最高,达到97.5%,查准率达到100%。对比实验结果表明,经该算法处理后,各模式识别精度均有所提升,其中,上楼模式查全率提升4.3%;行走模式查全率提升2.1%;下楼模式查全率提升4.7%,验证了降噪算法的有效性。

Aiming at the problem of poor data quality caused by random errors in microelectromechanical sensor accelerometers,which results in low accuracy of pedestrian activity recognition,a data noise reduction method for pedestrian activity recognition based on ensemble empirical mode decomposition algorithm is proposed.The accelerometer is used to collect training data and activity recognition data,and the ensemble empirical mode decomposition algorithm is used to decompose the acceleration signal into multiple eigenmode components and a residual component.The high frequency part of the modal component is removed and the remaining components are reconstructed.After the noise reduction signal is obtained,the time domain and frequency domain features are extracted,and model training and activity recognition are performed respectively.The results of three groups of pedestrian activity recognition experiments show that the data noise reduction pedestrian activity recognition method based on the ensemble empirical mode decomposition algorithm has a high recognition rate for the five daily pedestrian activities,with an average recall rate of 91.8%and an average precision rate of 92.3%,Among them,the highest recall rate of jogging reaches 97.5%,and the precision rate reaches 100%.Comparative experiment results show that the accuracy of each pattern recognition has been improved after the algorithm is processed.Among them,the recall rate of the upstairs mode is increased by 4.3%;the recall rate of the walking mode is increased by 2.1%;the recall rate of the downstairs mode is increased by 4.7%,verify the effectiveness of the noise reduction algorithm.