A Lightweight Passive Human Tracking Method Using Wi-Fi

Jian Fang, Lei Wang, Zhenquan Qin, Bingxian Lu, Wenbo Zhao, Yixuan Hou, Jenhui Chen2022 · DOI

TL;DR

This paper introduces WIDE, a lightweight, device-free human tracking system that leverages existing Wi-Fi infrastructure. By analyzing Channel State Information (CSI), WIDE can accurately track human movement in near real-time without requiring any special hardware on the tracked individuals, addressing limitations of existing methods that are often heavy, complex, or require extensive training.

What Problem Does This Paper Address?

Accurate indoor human tracking is crucial for various applications, from security to smart environments. Traditional methods often rely on cameras or specialized sensors, which can be costly, intrusive, and suffer from blind spots. Wireless sensing, particularly using Wi-Fi, offers a promising alternative due to its ubiquitous nature. However, existing Wi-Fi-based tracking methods often face challenges:

  • High Overhead and Latency: Many methods require significant data collection, training, or complex processing, making them unsuitable for real-time applications.
  • Device-Free Limitations: While device-free approaches (DfL) are attractive, traditional RSSI-based methods are coarse-grained and prone to multipath effects. More advanced CSI-based methods can be resource-intensive or require scenario-specific calibration.
  • Environmental Sensitivity: Indoor environments are complex, with reflections and signal fluctuations that can lead to errors. Methods need to be robust to these variations and ideally adapt to different environments without extensive recalibration.

This paper aims to develop a lightweight, device-free tracking method that is accurate, robust, and adaptable to various indoor settings by utilizing the existing Wi-Fi infrastructure.

Methods

The core of the WIDE method lies in analyzing the Channel State Information (CSI), which captures the wireless channel's characteristics between Wi-Fi transceivers. The authors propose a multi-stage approach:

  1. CSI Feature Extraction: They extract both amplitude and phase difference from CSI data. While amplitude is commonly used, the phase difference is highlighted as being less susceptible to clock synchronization errors and more sensitive to human activity.
  2. Environmental Adaptive Mechanism: To handle varying indoor environments, WIDE employs an adaptive mechanism. It first collects background CSI data in a static environment to establish a baseline. This baseline is then continuously updated to adapt to environmental changes, allowing the system to distinguish between normal environmental fluctuations and actual human movement. This is achieved by estimating the probability density function (PDF) of features from the static environment and setting adaptive thresholds.
  3. Anomaly Detection: A key component is the anomaly detection module. It calculates an "anomaly score" based on how much the current CSI features deviate from the established static environment profile. By integrating scores from multiple Wi-Fi links, WIDE improves detection accuracy and reduces false positives caused by general indoor noise.
  4. Wi-Fi Link Grid: WIDE models the indoor space as a grid of Wi-Fi links (formed by transceivers). By detecting when a target crosses these links in a temporal sequence, the system can reconstruct the target's trajectory. The method assumes a continuous trajectory and that a target occupies only one link at a time.
  5. Lightweight Processing: The method emphasizes efficiency by employing techniques like dimensionality reduction (using PCA on subcarriers) and careful feature selection to minimize computational overhead, enabling near real-time tracking.

Datasets

The paper describes experiments conducted in various indoor environments, including a conference room, office, and bedroom corridor. The data was collected using two Lenovo laptops equipped with Intel 5300 NICs as transceiver devices. The authors do not explicitly mention using pre-existing public datasets for training or evaluation; instead, they appear to have collected their own experimental data.

Key Insights and Ideas

  • Phase Difference as a Sensitive Feature: The paper emphasizes that the phase difference in CSI signals is a more robust and sensitive indicator of human activity compared to amplitude alone, especially when dealing with clock synchronization issues.
  • Environmental Adaptability is Crucial: A static threshold is insufficient for reliable tracking across different indoor settings. The proposed adaptive mechanism, which continuously updates a baseline based on environmental changes, is key to WIDE's robustness.
  • Grid-Based Tracking for Simplicity: Modeling the environment as a grid of Wi-Fi links simplifies the tracking problem. By detecting link crossings, the system can infer movement patterns without complex spatial mapping.
  • Lightweight Design Trade-off: WIDE prioritizes a lightweight design, accepting a slightly lower accuracy compared to some state-of-the-art methods in exchange for reduced computational complexity and easier scalability. This makes it practical for deployment on commodity hardware.

Key Results

  • Near Real-Time Tracking: The WIDE method achieves near real-time tracking with a window size of 0.5 seconds, indicating a low delay in detecting target movement.
  • Accuracy: The system demonstrates good performance, with True Positive (TP) rates reaching up to 91.4% and True Negative (TN) rates up to 95.2% when transceivers are 5 meters apart.
  • Robustness: The method shows robustness across different movement speeds and both Line-of-Sight (LoS) and Non-Line-of-Sight (nLoS) conditions.
  • Error Analysis: The theoretical average tracking error for a 3x3 transceiver setup in a 50 m² room is estimated to be around 0.95 meters. The study also shows that the tracking error grows slowly with increasing area.
  • Distance Influence: The accuracy drops significantly when the distance between transceivers exceeds 6 meters, suggesting an optimal deployment range.

Limitations and Future Work

  • Multi-Target Tracking Complexity: While the paper acknowledges the challenge of multi-target tracking, its current method is primarily designed for single targets. Distinguishing between multiple individuals, especially those with similar movement patterns or starting points, remains a significant challenge.
  • Accuracy vs. Overhead Trade-off: The authors explicitly state that WIDE's accuracy is slightly inferior to some existing methods but argue that this is a necessary trade-off for its lightweight nature and scalability.
  • Future Work:
    • Multiple Targets: Developing more sophisticated methods to handle simultaneous occurrences of multiple targets.
  • 3D Environments: Extending the tracking capabilities to three-dimensional spaces, such as for drone detection.
  • Long-Range Tracking: Combining WIDE with technologies like LoRa to enable outdoor, long-range tracking for IoT applications.
  • Enhanced Sensing: Investigating the integration of WIDE with signals possessing higher sensing abilities to improve performance and applicability in more scenarios.

Summary updated

BibTeX
@misc{fang2022lightweight,
  author = {Jian Fang and Lei Wang and Zhenquan Qin and Bingxian Lu and Wenbo Zhao and Yixuan Hou and Jenhui Chen},
  title = {{A Lightweight Passive Human Tracking Method Using Wi-Fi}},
  year = {2022},
  doi = {10.3390/s22020541},
}
Figure