Abstract:

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.