TL;DR
This paper introduces a novel, low-infrastructure method to track human queues in real-time using only smartphone WiFi signals and a single monitor. By analyzing unique patterns in signal strength, the system can accurately infer waiting and service times, opening doors for optimizing various service-oriented processes.
What Problem Does This Paper Address?
Monitoring human queues is crucial for optimizing operations in diverse settings like retail stores, airports, and hospitals. Traditional methods often rely on expensive, multi-point infrastructure like cameras or specialized sensors, which can be costly to install and maintain, and may raise privacy concerns. Existing wireless approaches, like using cellular data, are typically too coarse-grained to capture the dynamics of individual queues. This paper addresses the need for a more accessible, privacy-preserving, and fine-grained solution for real-time queue monitoring.
Methods
The core of this paper's contribution lies in its "feature-driven scheme" to identify critical time points in a queue: Beginning of Service (BoS), Leaving Point (LP), and End of Leaving (EoL). The approach leverages a single WiFi monitor placed near the service area, which captures the Received Signal Strength (RSS) from transmitting smartphones.
The system works by analyzing the RSS trace in a time-reversed manner, focusing on patterns associated with a person leaving the service point. Three key features are extracted from the RSS trace to identify the EoL:
- Longest consecutive negative-slope segments: Indicative of a rapid signal drop.
- Stable RSS with highest amplitude before leaving: Suggests the person was at the service point.
- Largest RSS decrease during the leaving period: A direct indicator of departure.
An "EoL Estimator" uses a weighted utility function (Equation 1) combining these features to pinpoint the EoL. Subsequently, a "LP/BoS Estimator" uses the KL-divergence technique to identify the LP and BoS. This involves comparing the distribution of RSS signal changes (slopes) before and after potential time points to find the point that maximally differentiates these distributions. This allows for the inference of waiting, service, and leaving periods.
Datasets
The authors conducted experiments in a laboratory setting using a custom setup. They collected 90 traces of human queues, with participants using an Android app to send beacon packets at a rate of 10 packets/sec. These traces were generated under three different service times: 30s, 60s, and 180s. The evaluation used a two-fold cross-validation approach, comparing the system's inferred time points against manually logged ground truth.
Key Insights and Ideas
- WiFi RSS is surprisingly informative: The paper demonstrates that the subtle fluctuations in WiFi signal strength, influenced by a person's movement relative to a monitor, contain enough information to precisely track queue dynamics.
- Time-reversed analysis is effective: By focusing on the most dramatic signal changes (leaving the service point), the system can reliably identify key transition points.
- Minimal infrastructure is sufficient: The success of a single WiFi monitor challenges the need for complex, multi-sensor systems, significantly reducing deployment costs and complexity.
- Feature engineering is key: The identification of specific RSS trace features (slopes, stability, drop magnitude) is crucial for robustly inferring queue events.
Key Results
The system demonstrates robustness and accuracy in measuring queue parameters.
- The average error in estimating critical time points (LP and BoS) was less than 10 seconds for short (30s) and normal (60s) service times, even with a single antenna.
- Using two antennas at the monitor improved performance, especially for longer service times (180s).
- The average error for measuring waiting and service times was also less than 10 seconds for short and normal service times.
- For longer service times, the two-antenna approach showed over 30% improvement compared to the single-antenna approach, though both experienced some performance degradation.
Limitations and Future Work
- Laboratory setting: The experiments were conducted in a controlled laboratory environment. Real-world deployments may face more complex environmental factors (e.g., multipath interference, signal fading due to more dynamic environments).
- Single monitor assumption: While the paper advocates for a single monitor, its placement and effectiveness might be sensitive to the physical layout of the service area.
- Privacy: While less intrusive than cameras, the use of WiFi signals still raises questions about user privacy that might need further consideration.
Future work could involve testing the system in diverse real-world scenarios, exploring the impact of different WiFi monitor placements, and investigating methods to further enhance accuracy and robustness in challenging environments. The authors also suggest that the extracted queue metrics could be used to build more sophisticated predictive models for queue behavior.
Summary updated