Super Low Resolution RF Powered Accelerometers for Alerting on Hospitalized Patient Bed Exits

Michael Chesser, Asangi Jayatilaka, Renuka Visvanathan, Christophe Fumeaux, Alanson P. Sample, Damith Chinthana Ranasinghe

This paper presents a clever, low-cost solution to a high-stakes problem in healthcare: patient falls. By rethinking what an "accelerometer" looks like, the researchers from the University of Adelaide and the University of Michigan have developed a batteryless system capable of detecting when a patient leaves their bed with high accuracy.

TL;DR

Researchers developed the "ID-Sensor," a batteryless RFID tag that functions as a 1-bit, "super low resolution" accelerometer by toggling between two unique IDs via a mechanical tilt switch. Using a Fully Convolutional Network (FCN) to process raw RFID data, the system detects hospital bed exits with a 90% F-score, offering a maintenance-free and privacy-preserving alternative to cameras or battery-powered wearables.

What Problem Does This Paper Address?

Falls in hospitals and nursing homes are a massive burden, costing the US healthcare system roughly $50 billion annually. Most falls occur when unassisted patients attempt to get out of bed.

Current solutions are riddled with trade-offs:

  • Pressure Mats: Often yield high false-alarm rates and require constant maintenance/disinfection.
  • Cameras: Raise significant privacy concerns for patients.
  • Battery-Powered Wearables: These are expensive, bulky, and require frequent charging or battery replacements—a logistical nightmare in a busy hospital ward.

The authors identify a key insight: human activity occurs at very low frequencies (typically < 4 Hz). We don't need high-fidelity, power-hungry MEMS accelerometers to detect a bed exit; we just need enough resolution to capture the "macro" movements of a patient sitting up and standing.

Methods

The core innovation is the ID-Sensor. It consists of two commercial off-the-shelf (COTS) RFID chips connected to a single antenna through a mechanical tilt switch (a small steel ball in a tube).

  • The Hardware: As the patient moves, the steel ball rolls, connecting either "ID 1" or "ID 2" to the antenna. This effectively encodes physical orientation as a change in the reported digital identifier.
  • The Data: The system collects a 4-tuple of data: Reader Antenna ID, RSSI (signal strength), Phase, and the Tag ID.
  • The AI: The researchers compared traditional Logistic Regression (with engineered features) against two deep learning architectures:
  • ConvLSTM: Combines CNNs for feature extraction with LSTMs to capture temporal dependencies.
  • FCN (Fully Convolutional Network): A 1D convolutional architecture that treats the RFID stream as a multi-channel signal.

The FCN emerged as the winner, proving more convenient to train while effectively handling the "window label ambiguity" common in activity recognition.

Datasets

The study utilized a custom dataset (which the authors have released to the research community) involving 23 hospitalized older patients (average age ~84).

  • Setting: Real-world hospital rooms (single and double) at the Queen Elizabeth Hospital.
  • Scale: 234,764 tag readings and 70 ground-truth bed-exit events.
  • Significance: Unlike many studies that use healthy young volunteers in labs, this dataset captures the messy, "intra-class diversity" of actual frail patients in a clinical environment.

Key Insights and Ideas

  • ID as a Sensor: Traditionally, an RFID ID is just a static "name." Here, the toggling between IDs is treated as a primary data source, turning a simple identification protocol into a motion sensor.
  • Zero-Power Sensing: The "sensing" mechanism (the rolling ball) is entirely mechanical and consumes no electrical power. The energy to transmit the data is harvested from the RFID reader's signal.
  • FCN Efficiency: The FCN performed better than the more complex ConvLSTM. This suggests that the local patterns in RSSI and ID-toggling are highly discriminative for bed exits, and the long-term memory of an LSTM might not be strictly necessary for this specific task.
  • Privacy by Design: Because the system only sees signal metadata and IDs, it is inherently more private than video-based monitoring.

Key Results

The "ID-Sensor" approach significantly outperformed a standard "Tag" approach (which ignores the ID toggling).

  • Top Performance: The FCN model achieved an F-score of 0.90 (Precision: 0.85, Recall: 0.97).
  • Latency: Most bed exits were detected within 10 seconds of the actual event, providing enough lead time for nursing staff to intervene.
  • Robustness: The system maintained high performance across different room layouts (single vs. double beds), suggesting it is "environment agnostic."

Limitations and Future Work

  • Study Duration: The pilot trials lasted only 60–90 minutes per patient. Long-term reliability over days or weeks remains to be seen.
  • Mechanical Wear: The authors haven't yet evaluated the mean time to failure for the mechanical tilt switch.
  • Infrastructure: While the tags are dirt cheap (~$0.10), the system requires RFID reader infrastructure, which is currently an upfront cost for hospitals (though many are already installing it for asset tracking).
  • Next Steps: The team aims to integrate these sensors directly into the fabric of hospital gowns and explore using two sensors (one on each shoulder) to further reduce false positives caused by patients simply rolling over in their sleep.

Summary updated

Abstract

Falls have serious consequences and are prevalent in acute hospitals and nursing homes caring for older people. Most falls occur in bedrooms and near the bed. Technological interventions to mitigate the risk of falling aim to automatically monitor bed-exit events and subsequently alert healthcare personnel to provide timely supervisions. We observe that frequency-domain information related to patient activities exist predominantly in very low frequencies. Therefore, we recognise the potential to employ a low resolution acceleration sensing modality in contrast to powering and sensing with a conventional MEMS (Micro Electro Mechanical System) accelerometer. Consequently, we investigate a batteryless sensing modality with low cost wirelessly powered Radio Frequency Identification (RFID) technology with the potential for convenient integration into clothing, such as hospital gowns. We design and build a passive accelerometer-based RFID sensor embodiment-ID-Sensor-for our study. The sensor design allows deriving ultra low resolution acceleration data from the rate of change of unique RFID tag identifiers in accordance with the movement of a patient's upper body. We investigate two convolutional neural network architectures for learning from raw RFID-only data streams and compare performance with a traditional shallow classifier with engineered features. We evaluate performance with 23 hospitalized older patients. We demonstrate, for the first time and to the best of knowledge, that: i) the low resolution acceleration data embedded in the RF powered ID-Sensor data stream can provide a practicable method for activity recognition; and ii) highly discriminative features can be efficiently learned from the raw RFID-only data stream using a fully convolutional network architecture.
BibTeX
@inproceedings{chesser2019super,
  author = {Michael Chesser and Asangi Jayatilaka and Renuka Visvanathan and Christophe Fumeaux and Alanson P. Sample and Damith Chinthana Ranasinghe},
  title = {{Super Low Resolution RF Powered Accelerometers for Alerting on Hospitalized Patient Bed Exits}},
  booktitle = {PerCom},
  year = {2019},
  url = {https://openreview.net/forum?id=C6rFCLgDhVO1},
}

Loading related content...

Figure