Scholarlink

<p>Real-time monitoring of human and vehicle movement is increasingly critical for societal applications ranging from urban planning and traffic management to public safety and disaster response. While traditional monitoring relies on cameras and mobile signal data, these approaches face significant limitations regarding privacy, line-of-sight visibility, and spatial resolution. To address these challenges, this study proposes a privacy-preserving monitoring framework that uses ground-vibration data from a seismometer network to detect, classify, and track moving objects. For classification, a convolutional neural network (CNN) was trained on time–frequency spectrograms. By recording roadside vibrations from vehicles and pedestrians, as well as background noise, using seismometers, the model achieved a test accuracy of 98.6% across the three classes. Notably, it correctly identified nearly all vehicle-induced vibrations within 30&#xa0;m and human-induced vibrations within 10&#xa0;m. Misclassifications caused by signal attenuation at greater distances were effectively mitigated through a rule-based post-processing step that accounts for temporal continuity. For localization of these moving objects, two methods were evaluated: (1) a particle motion-based approach utilizing surface wave arrival directions, and (2) a grid search method based on Time-Difference-Of-Arrival (TDOA). Experimental results confirmed that both methods successfully tracked object trajectories; notably, the particle motion approach achieved a localization accuracy within several meters for sources located inside or near the sensor array. These findings demonstrate that seismic signal analysis offers a robust, non-intrusive alternative for monitoring pedestrian and vehicular traffic movements.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Privacy-preserving detection, classification, and tracking of human and vehicle movements using seismic sensor networks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月21日

查看全文
  • Yuki Oi,
  • Ahmad Bahaa Ahmad,
  • Takeshi Tsuji

摘要

Real-time monitoring of human and vehicle movement is increasingly critical for societal applications ranging from urban planning and traffic management to public safety and disaster response. While traditional monitoring relies on cameras and mobile signal data, these approaches face significant limitations regarding privacy, line-of-sight visibility, and spatial resolution. To address these challenges, this study proposes a privacy-preserving monitoring framework that uses ground-vibration data from a seismometer network to detect, classify, and track moving objects. For classification, a convolutional neural network (CNN) was trained on time–frequency spectrograms. By recording roadside vibrations from vehicles and pedestrians, as well as background noise, using seismometers, the model achieved a test accuracy of 98.6% across the three classes. Notably, it correctly identified nearly all vehicle-induced vibrations within 30 m and human-induced vibrations within 10 m. Misclassifications caused by signal attenuation at greater distances were effectively mitigated through a rule-based post-processing step that accounts for temporal continuity. For localization of these moving objects, two methods were evaluated: (1) a particle motion-based approach utilizing surface wave arrival directions, and (2) a grid search method based on Time-Difference-Of-Arrival (TDOA). Experimental results confirmed that both methods successfully tracked object trajectories; notably, the particle motion approach achieved a localization accuracy within several meters for sources located inside or near the sensor array. These findings demonstrate that seismic signal analysis offers a robust, non-intrusive alternative for monitoring pedestrian and vehicular traffic movements.