基于DAS信号和CNN分类算法的人员运动轨迹监测方法

Personnel Motion Trajectory Monitoring Method Based on DAS Signal and CNN Classification Algorithm

人员运动轨迹监测目前主要依靠摄像头网络、热成像和雷达系统等方式实现,但这些方法的监测范围有限,且易受到光线、热源、遮挡物和距离的影响.针对上述问题,我们开发了一种利用分布式声学传感(distributed acoustic sensing,DAS)识取人员振动信号用于监测人员运动轨迹的新方法,能够不受光线、热源、遮挡物等环境改变的影响,实现长距离连续监测.首先,通过长短时窗比值(STA/LTA)法自动拾取DAS信号,再对拾取的信号进行分析,将其分为人员运动、重锤和噪声3类信号;然后,构建卷积神经网络(CNN),将拾取的3类信号作为CNN数据库输入到网络中学习和训练;最后,通过实际数据测试得到3类信号分类结果.结果表明,分类结果的识别准确率均达到80.00%以上.利用识别得到的信号,通过CNN分类结果可确定人员所在的光纤道,追踪人员位置,输入更新的DAS信号连续识别人员运动信号,监测人员运动轨迹;然后用单位时间内识别人员移动的通道数计算运动速率.

At present,the monitoring methods of personnel motion trajectory mainly rely on camera networks and thermal imaging and radar systems.However,these methods have limited monitoring range and are easily affected by light,heat sources,obstructions and distance.In response to the above problems,we have developed a new method of using distributed acoustic sensing(DAS)to recognize human vibration signals to monitor personnel motion trajectory,which can be undisturbed and unaffected by environmental changes such as light,heat sources,and obstructions over long distances.First,the DAS signals are automatically picked up by the long-to-short time window ratio(STA/LTA)method,and then the picked-up signals are analyzed and can be divided into three types:personnel motion,hammer,and noise.Further,a convolutional neural network(CNN)architecture is constructed,and the three types of signals are input into the network as a CNN database for learning and training.Finally,the results of the three types of signal classification results are obtained through the actual data test.The accuracy reaches more than 80.00%.The recognized signal by CNN classification result can be used to determine the person’s fiber channel to track the position of the person,the personnel motion trajectory can be monitored by input the updated DAS signal continuously to identify the personnel motion signal,and the motion speed can be calculated by the number of personnel movement channels identified per unit time.