Accurate Device-Free Tracking Using Inexpensive RFIDs

Liyao Li, Chongzheng Guo, Yang Liu, Lichao Zhang, Xiaofei Qi, Yuhui Ren, Baoying Liu, Feng Chen2018 · DOI

TL;DR

This paper introduces RDTrack, a novel device-free tracking system that uses inexpensive RFID tags to accurately track human movement. By analyzing Doppler shifts, which are less affected by environmental multipath, RDTrack achieves a low tracking error of 32 cm, making it robust across various indoor environments.

What Problem Does This Paper Address?

Device-free tracking, which allows for monitoring without requiring targets to carry any active devices, is crucial for applications like smart homes, security, and fitness. Existing methods often rely on Received Signal Strength Indicator (RSSI) or Channel State Information (CSI) from Wi-Fi or other RF devices. However, these approaches are highly susceptible to environmental changes, particularly multipath reflections, leading to poor performance in diverse settings. Furthermore, many current systems require extensive training data, mass deployment of expensive hardware, or constant recalibration. The challenge is to develop a robust, accurate, and cost-effective device-free tracking system that can adapt to different environments without prior training.

Methods

RDTrack tackles this challenge by leveraging the Doppler frequency shift, which is less sensitive to multipath interference. The core of the system involves several key steps:

  • Data Extraction and Revision: Raw RFID signals are processed to extract phase values. A novel "self-adaption dynamic protocol" is introduced to address phase ambiguity and random hopping issues.
  • Noise Reduction using DWT: The Discrete Wavelet Transform (DWT) is employed to denoise the phase values. Specifically, the "heursure" threshold method is used to effectively remove noise while preserving important signal features.
  • Doppler Shift Calculation: Cleaned phase values are used to calculate accurate Doppler shifts, which are then used to create "Doppler profiles" representing different motion patterns.
  • Motion Detection and Segmentation: DWT is used again to identify the beginning and end of the time intervals when tags are affected by target motion, effectively segmenting the data.
  • Trajectory Estimation using HMM: A Hidden Markov Model (HMM) is utilized to estimate the fine-grained trajectory. This model identifies "Path Information Units" (PIUs) representing distinct motion segments (e.g., going straight, turning).
  • Trajectory Smoothing: Locally-weighted linear regression is applied to smooth the estimated trajectory, ensuring a more accurate and continuous path.

Datasets

The paper does not explicitly mention pre-existing datasets for training or evaluation. Instead, the authors collected their own data by conducting experiments in four different real-world environments: a corridor, a library reading room, a laboratory, and a typical apartment. They generated over 1000 "primitive data profiles" by having volunteers perform various paths and motions.

Key Insights and Ideas

  • Doppler Shift Robustness: The paper highlights that Doppler shifts, derived from phase changes, are significantly more robust to multipath interference than RSSI or CSI, making them ideal for device-free tracking in complex environments.
  • DWT for Motion Segmentation: The innovative use of DWT to precisely detect the start and end times of tag influence due to motion is a key contribution, enabling accurate segmentation without relying on explicit motion sensors.
  • HMM for Fine-Grained Trajectory: Employing an HMM for trajectory estimation allows for the identification of distinct motion patterns (PIUs), leading to a more detailed and accurate representation of the target's path.
  • Adaptability without Training: A significant advantage of RDTrack is its ability to perform well in different environments without requiring prior training data specific to each location, a common limitation in many existing systems.

Key Results

  • Low Tracking Error: RDTrack achieves a low tracking error of 32 cm. This error can be further reduced to 27 cm when using two RFID systems.
  • High Segmentation Accuracy: The segmentation accuracy of RDTrack is reported to be between 96.9% and 100%, with an average accuracy of 98.0% for "going straight," 97.42% for "going backward," and 97.0% for "turning."
  • Tag Discovery Accuracy: RDTrack demonstrates a 100% tag discovery accuracy, outperforming existing methods like Rollcaller (95%) and Twins (98.8%).
  • Motion Recognition: The system achieves a True Positive Rate (TPR) between 82.6% and 94.29% and a False Positive Rate (FPR) less than 13% across different environments and motions.
  • System Performance: 80% of anchor tags are measured within a 30-cm distance drift, indicating good localization precision. The average position error is 0.33 m, significantly better than comparable methods.

Limitations and Future Work

  • Human Diversity Impact: While the system is generally robust, the paper notes that variations in human body shape can slightly affect accuracy. Increasing the fingerprint database with more volunteer data can mitigate this.
  • Motion Speed: The accuracy of motion recognition is stable for speeds up to 2.5 m/s. However, at higher speeds, the TPR decreases, and the FPR increases, suggesting that very rapid movements might pose a challenge.
  • Number of Tags: The system performs best with approximately 13 tags. While increasing tags initially improves accuracy, exceeding this number can lead to a decrease in performance.
  • Environmental Challenges: In environments with numerous barriers or less effective detection areas (like a narrow corridor), accuracy can be lower due to signal blockage and reduced detection range.
  • Future Work: The authors are currently extending the system to standard Wi-Fi, facing the challenge of extracting clean phase values from Wi-Fi signals. They also aim to further improve the system's robustness and accuracy.

Summary updated

BibTeX
@misc{li2018accurate,
  author = {Liyao Li and Chongzheng Guo and Yang Liu and Lichao Zhang and Xiaofei Qi and Yuhui Ren and Baoying Liu and Feng Chen},
  title = {{Accurate Device-Free Tracking Using Inexpensive RFIDs}},
  year = {2018},
  doi = {10.3390/s18092816},
}
Figure