⚡ electrical engineering

Floor Plan-Agnostic Detection of Gait Speed Drifts Using Ambient Sensors

This paper proposes a novel, floor plan-agnostic method that uses sparse ambient sensors to detect gait speed drifts in older adults by analyzing sensor-to-sensor transition durations, achieving performance comparable to or exceeding state-of-the-art baselines that require detailed home layouts.

Original authors: Marina Vicini, Martin Rudorfer, Zhuangzhuang Dai, Ahmad Beltagui, Luis J. Manso

Published 2026-05-05

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Marina Vicini, Martin Rudorfer, Zhuangzhuang Dai, Ahmad Beltagui, Luis J. Manso

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine your home is a busy city, and you are a traveler walking through its streets every day. Usually, you walk at a steady pace. But sometimes, due to health issues, your walking speed might slowly start to slow down. Detecting this "drift" in speed early is like spotting a crack in a dam before it breaks—it can help prevent bigger problems later.

Traditionally, to measure how fast you walk, doctors need to see you in a clinic or install a very specific, expensive line of sensors in your hallway, like a toll booth that only works if you know the exact map of the road. But what if you don't have the map? What if you just have a few cheap, privacy-friendly motion sensors scattered around your house, like streetlights in a city you've never seen a map of?

This paper proposes a clever new way to solve that puzzle. Here is how it works, broken down into simple concepts:

The Problem: The "Blind" City

Most smart home systems that try to measure walking speed need a floor plan (a map of the house). They need to know exactly where the walls are and how far apart the sensors are to calculate speed. But getting a floor plan is often impossible or too expensive for a typical home. Without a map, the sensors are like blindfolded people trying to guess how fast a car is driving just by hearing it pass by.

The Solution: Listening to the "Beeps"

The researchers created a method that doesn't need a map. Instead, it treats the sensors like a series of bells placed around the house.

The Bell Ring: When you walk past a sensor, it rings (turns ON). When you leave, it stops (turns OFF).
The Handoff: The system looks for moments when one bell rings, and then a different bell rings shortly after. This is a "handoff"—a sign you moved from one spot to another.
The Stopwatch: It measures exactly how long it took to get from Bell A to Bell B.

The Filter: Ignoring the Noise

Since the system doesn't know the map, it can't tell if you actually walked or just waved your hand in front of two sensors that are right next to each other.

The "Too Fast" Filter: If the time between bells is less than a second, the system assumes you didn't really walk; you probably just triggered two sensors that overlap. It ignores these.
The "Too Slow" Filter: If it takes a minute to get from one sensor to another, you probably stopped to make coffee or talk to someone. The system ignores these too.

The Detective Work: Finding the Drift

Once the system has a list of valid "walks" (the time it took to go from sensor to sensor), it acts like a detective looking for changes in your habits.

The Baseline: It looks at your walking times from the first week (the "old you").
The Check-up: It looks at your walking times from the last week (the "new you").
The Comparison: It uses a statistical test (a math tool that compares two groups of numbers) to ask: "Is the new group of times significantly slower than the old group?"

If the answer is yes, the system raises an alert: "Your walking speed has drifted."

The Results: How Well Did It Work?

The researchers tested this idea using a computer simulation of four different apartment layouts (like four different puzzle pieces). They created a "virtual person" who walked at a normal speed for 100 days, then suddenly started walking slower.

The Competition: They compared their "no-map" method against a high-tech "map-required" method (the current gold standard).
The Outcome: Surprisingly, the "no-map" method performed just as well, and in some apartment layouts, it was even better than the method that needed the map.
The Catch: It was slightly slower at spotting tiny changes in speed (like a very small slowdown), but it was very good at spotting the bigger, clinically important changes.

The Bottom Line

This paper proves that you don't need a blueprint of someone's house to monitor their walking speed. By simply listening to the timing of when cheap motion sensors go off, you can detect if an older adult is slowing down. This makes it possible to add health monitoring to existing, affordable smart home kits without needing expensive new hardware or invasive cameras.

Important Note: The paper tested this on a computer simulation, not real people in real homes yet. The "virtual person" in the test walked perfectly straight and didn't have visitors. The authors admit that real life is messier, and future work is needed to see if this works when real people are walking around with all their daily distractions.

Technical Summary: Floor Plan-Agnostic Detection of Gait Speed Drifts Using Ambient Sensors

Problem Statement

Gait speed is a critical health indicator for older adults, where a decline of 0.10 m/s is considered a minimal clinically important difference reflecting functional decline. While continuous home monitoring offers the potential for early diagnosis, existing ambient sensor-based methods for gait speed estimation typically require detailed knowledge of the home's floor plan or specific, linear sensor configurations. Acquiring floor plans is often economically unfeasible, and specialized sensor setups hinder scalability and cost-effectiveness. Consequently, there is a lack of methods capable of detecting gait speed drifts using sparse, commercially available ambient sensors (e.g., Passive Infrared (PIR) and door sensors) without prior spatial context.

Methodology

The proposed approach detects gait speed drifts by analyzing the duration of transitions between sensor activations, operating entirely without knowledge of the home layout, sensor locations, or walking paths.

Data Representation: The system ingests a raw stream of sensor events (ON/OFF signals) from PIR and door sensors. It constructs sequences of transitions between distinct sensors ( $s_i \to s_j$ ), recording the time difference ( $\delta$ ) between activations and the day of occurrence ( $\gamma$ ).
Filtering: To isolate valid ambulation from stationary or overlapping sensor noise, two filters are applied:
- Minimum Duration ( $T_{min}$ ): Excludes transitions with near-zero duration (e.g., $<1$ second), which likely result from overlapping sensor fields rather than walking.
- Maximum Duration ( $T_{max}$ ): Excludes excessively long sequences (e.g., $>60$ seconds) that likely represent multiple interwoven activities rather than a continuous path.
Feature Extraction: For each unique sensor pair, the system aggregates transition durations for a specific day. It computes a specific percentile ( $p_k$ ) of these durations (e.g., the 0th percentile, or minimum duration) to serve as a robust daily mobility metric. A minimum support threshold ( $S_{min}$ ) ensures sufficient data points exist for statistical validity.
Drift Detection: The method compares a recent "query period" (e.g., the last 7 days) against a stable "reference period" (e.g., the first 7 days). For each sensor pair, a non-parametric Mann-Whitney U test determines if the distribution of transition durations in the query period differs significantly from the reference.
Aggregation: Individual sequence-level test results are aggregated into a single daily drift decision. The paper evaluates both simple aggregation and a frequency-weighted ensemble, where decisions are weighted by the number of times a specific sensor transition was observed.

Key Contributions

The paper presents three primary contributions:

Problem Formulation: A formal definition for detecting gait speed drifts using sparse ambient sensor data without requiring spatial context.
Simulated Dataset: A synthetic dataset generated using an agent-based behavioral simulator, featuring four distinct studio apartment layouts and a wide range of gait speeds (0.4 m/s to 1.2 m/s) with simulated sudden drifts.
Novel Method: A floor plan-agnostic detection algorithm that identifies informative sensor-to-sensor transitions and analyzes their duration fluctuations to detect drifts, eliminating the need for additional spatial relations or hardware.

Experimental Results

The method was evaluated on a simulated dataset across four different home layouts and compared against a state-of-the-art baseline (Nait Aicha et al.) that relies on floor plan information to identify optimal walking paths.

Performance Metrics: The proposed method achieved an average F1-Score of 0.75 (standard deviation 0.25) and an accuracy of 0.78, outperforming the baseline (F1-Score 0.72, Accuracy 0.69) in Layouts A and B. However, it underperformed slightly in Layouts C and D, where the baseline's knowledge of optimal paths allowed for earlier detection (lower detection delay).
Parameter Sensitivity:
- Percentile: The 0th percentile (minimum duration) yielded the highest performance in the clean simulated environment.
- Filters: A minimum duration filter ( $T_{min} = 1$ second) was critical for removing non-informative sequences caused by overlapping sensor fields. The maximum duration filter had negligible impact in the simulation.
- Support: A minimum support threshold of $S_{min} = 0$ (no threshold) provided the best results; higher thresholds excluded too much data.
- Weighting: A simple, unweighted ensemble performed slightly better than a frequency-weighted one.
Drift Magnitude: Detection sensitivity decreased as the magnitude of the speed gap decreased. While the baseline showed an advantage at very small gaps (0.1–0.2 m/s), the proposed method remained competitive, demonstrating utility even for subtle changes.

Significance and Limitations

The paper claims that this work demonstrates a feasible, scalable, and cost-effective approach for gait drift monitoring that can be retrofitted to existing commercial smart home systems without requiring new hardware or floor plans. By achieving performance comparable to floor plan-dependent methods, it offers a pathway for widespread deployment in real-world settings where spatial data is unavailable.

The authors explicitly note the following limitations:

Synthetic Data: Evaluation was conducted solely on simulated data, which lacks the noise, multi-resident complexity, and trajectory variability of real-world environments.
Sensor Specificity: The study focused on PIR sensors with a 1.0 m radius; the impact of varied sensor ranges or types was not fully explored.
Drift Types: The current method is designed to detect sudden drifts against a fixed initial baseline. It is not yet optimized for detecting gradual, recurrent, or incremental declines, nor does it address the challenge of data association in multi-resident households.
Reference Period: The fixed reference period (initial week) may not be robust if it inadvertently contains drifts; dynamic or rolling windows are suggested for future work.