Activity recognition based on micro-Doppler signature with in-home Wi-Fi

Qingchao Chen, Bo Tan, Kevin Chetty, Karl Woodbridge

Here is a summary of the paper:

TL;DR

This paper introduces a contactless, privacy-preserving system for recognizing human activities in the home using ambient Wi-Fi signals. By analyzing micro-Doppler signatures with a Sparse Representation Classifier (SRC), the system can instantly identify critical events like falls without requiring users to wear sensors or install cameras.

What Problem Does This Paper Address?

As the global population ages, monitoring the elderly or disabled in their homes (Ambient Assisted Living) is becoming critical. Traditional monitoring relies on wearables (which users often forget or refuse to wear due to discomfort), cameras (which violate privacy and require optimal lighting), or PIR/RFID sensors (which offer low resolution or require complex, expensive setups).

The authors tackle the need for an unobtrusive, device-free way to reliably detect specific human movements—especially life-threatening ones like falls—in real-time. Crucially, healthcare applications require "one-shot" classification; the system must recognize an activity from a single motion cycle to trigger an instant alert, rather than relying on continuous, repeated movements (like tracking a walking gait over several seconds).

Methods

The authors built a passive Wi-Fi radar system using Software Defined Radios (SDRs) that leverages existing in-home Wi-Fi access points as transmitters of opportunity. The pipeline consists of three main steps:

  1. Signal Extraction: They use the Cross Ambiguity Function (CAF) to extract micro-Doppler (m-D) signatures (time-velocity patterns) from Wi-Fi signals reflecting off the human body.
  2. Data Alignment: Because human motions take varying amounts of time, they developed an automatic start/end point detection method using the weighted standard deviation of frequency bins. They then use bi-cubic interpolation to resize all captured m-D signatures into uniform matrices.
  3. Classification: Instead of manually extracting empirical features (like maximum Doppler shift), they apply Principal Component Analysis (PCA) for dimensionality reduction, followed by a Sparse Representation Classifier (SRC). The SRC uses a Subspace Pursuit algorithm to find the sparsest linear combination of training samples that can reconstruct the test sample. This is compared against a standard Support Vector Machine (SVM) baseline.

Datasets

The authors collected a custom dataset in a standard living room environment (3.87m x 3.67m) using three synchronized USRP N210 SDRs (one reference channel, two surveillance channels). Four male subjects performed six specific activities at a fixed location:

  1. Picking an object up and standing
  2. Sitting down on a chair
  3. Standing up from a chair
  4. Falling down onto a mattress
  5. Standing up after falling
  6. Lying on a mattress and getting out of it

The dataset consists of 170 total motion samples per receiver channel.

Key Insights and Ideas

  • Embracing Interference: In traditional radar, Direct Signal Interference (DSI) and ambiguity peaks are usually filtered out as noise. Here, the authors intentionally keep them. They realized that the presence or absence of a zero-Doppler line (e.g., caused by a human body physically blocking the line-of-sight to the receiver during a specific motion) acts as a highly discriminative feature for classification.
  • Sparsity Beats Margins: The SRC vastly outperformed the traditional SVM. SVM struggled with inter-class similarity (different motions that produce similar Doppler shifts). SRC succeeded because it strictly selects the most compact, sparse basis to represent the test sample, naturally ignoring misleading boundary samples between similar classes.
  • Overcoming Wi-Fi's Physical Limits: Wi-Fi operates at 2.4 GHz, meaning the maximum human walking speed (~2 m/s) only induces a tiny Doppler shift of about 32 Hz. Despite this narrow frequency range, the passive radar integrates enough signal samples to create distinct, recognizable time-velocity patterns for different activities.

Key Results

  • The SRC achieved an average classification accuracy of 90.2% on Channel 1 and 85.2% on Channel 2.
  • In stark contrast, the baseline SVM classifier only achieved 46% accuracy across both channels.
  • The system proved highly sensitive to receiver geometry. For example, Channel 2 (a bistatic setup) was much better at detecting the "falling down" motion than Channel 1 (a quasi-monostatic setup) because the falling motion induced a stronger Doppler frequency relative to Channel 2's physical position in the room.

Limitations and Future Work

Limitations: The system's accuracy is highly dependent on the physical angle and position of the receiver relative to the subject. Additionally, the dataset is relatively small and lacks demographic diversity (only four male subjects), meaning the model's generalization to different body types or movement speeds is untested.

Future Work: To solve the geometry problem, the authors propose moving toward "joint sparsity-based classifiers." By fusing data from multiple Wi-Fi receiver channels simultaneously—and eventually integrating other modalities like wearable accelerometers and video—the system could overcome the blind spots of any single receiver, increasing overall coverage and robustness in real-world homes.

Summary updated

Abstract

Device free activity recognition and monitoring has become a promising research area with increasing public interest in pattern of life monitoring and chronic health conditions. This paper proposes a novel framework for in-home Wi-Fi signal-based activity recognition in e-healthcare applications using passive micro-Doppler (m-D) signature classification. The framework includes signal modeling, Doppler extraction and m-D classification. A data collection campaign was designed to verify the framework where six m-D signatures corresponding to typical daily activities are sucessfully detected and classified using our software defined radio (SDR) demo system. Analysis of the data focussed on potential discriminative characteristics, such as maximum Doppler frequency and time duration of activity. Finally, a sparsity induced classifier is applied for adaptting the method in healthcare application scenarios and the results are compared with those from the well-known Support Vector Machine (SVM) method.
BibTeX
@inproceedings{chen2016activity,
  author = {Qingchao Chen and Bo Tan and Kevin Chetty and Karl Woodbridge},
  title = {{Activity recognition based on micro-Doppler signature with in-home Wi-Fi}},
  booktitle = {HealthCom},
  year = {2016},
  url = {https://openreview.net/forum?id=MveLiY_fMm},
}

Loading related content...

Figure