LoGait: LoRa Sensing System of Human Gait Recognition Using Dynamic Time Warping
LoGait: LoRa Sensing System of Human Gait Recognition Using Dynamic Time Warping
复制标题
DOI：
10.1109/jsen.2023.3297438
复制
发表时间：
2023-09
影响因子：
4.3
通讯作者：
Y. Ge;Wenda Li;Muhammad Farooq;Adnan Qayyum;Jingyan Wang;Zikang Chen;Jonathan Cooper;M. Imran;Q. Abbasi
中科院分区：
文献类型：
--
作者：
Y. Ge;Wenda Li;Muhammad Farooq;Adnan Qayyum;Jingyan Wang;Zikang Chen;Jonathan Cooper;M. Imran;Q. Abbasi
基于视觉的步态分析和人体识别系统已在文献中被广泛提出。然而，由于视频质量、遮挡和严重的隐私问题等方面的挑战，这些系统不能很容易地应用于许多实时应用。为了克服这些问题，我们提出了利用无处不在的LoRa信号识别不同室内环境中的步态的LoGait系统。我们的工作是基于直觉，即不同用户的行走模式可以通过不同的步幅和频率来区分。受人体行走干扰的无线LoRa信号捕捉被测者的步态信息。结合LoRa信号的远距离传输能力，与基于wifi的步态识别系统相比，该系统可以实现更大的步态识别感知范围。提出的LoGait系统利用两个LoRa接收器信道之间的相位差，以及一组滤波技术，提取不同的特征并生成人体步态轮廓。然后使用基于动态时间扭曲（DTW）的识别算法将该轮廓与数据库进行匹配，从而实现基于独特步态模式的准确识别。在视线（LOS）、非视线（NLOS）和远距离三种不同的步态识别场景下进行了验证，准确率分别为85.13%、79.14%和84.14%。
Vision-based gait analysis and human identification systems have been widely proposed in the literature. However, these systems cannot be readily applied in many real-time applications due to involved challenges such as video quality, occlusion, and serious privacy concerns. To overcome such issues, we propose the LoGait system that leverages ubiquitous LoRa signals recognize gait in different indoor environments. Our work is based on the intuition that the walking pattern of different users can be distinguished by distinct stride size and frequency. The wireless LoRa signal which is interfered by human walking will capture the gait information of subjects. In combination with the long-distance transmission ability of LoRa signal, the system enables a larger sensing range of gait recognition compared to the WiFi-based gait recognition system. The proposed LoGait system utilizes the phase difference between two LoRa receiver channels, along with a set of filtering techniques, to extract distinctive features and generate a human gait profile. This profile is then matched against a database using a dynamic time warping (DTW)-based recognition algorithm, enabling accurate identification based on unique gait patterns. It has been validated in three different scenarios for gait recognition namely line of sight (LOS), non-LOS (NLOS), and long distance, with an accuracy of 85.13%, 79.14%, and 84.14%, respectively.