Exploring the Use of Ambient WiFi Signals to Find Vacant Houses

Shin'ichi Konomi, Tomoyo Sasao, Simo Hosio, Kaoru SezakiAmI 2017 · DOI

TL;DR

This paper proposes a novel method to identify occupied houses using ambient WiFi signals collected via opportunistic smartphone sensing. By constructing a hybrid network of real and virtual WiFi access points and analyzing their spatial patterns, the technique can significantly reduce the cost and effort of traditional field surveys for detecting vacant properties.

What Problem Does This Paper Address?

Many countries, particularly Japan, are facing a growing problem of vacant houses due to population decline and urban concentration. This issue leads to decreased land quality, increased crime and disaster risks, and a decline in community vitality. Traditionally, identifying these vacant houses requires costly and time-consuming manual field surveys. The authors aim to address this by developing a more efficient, technology-driven approach to pinpoint occupied homes, thereby indirectly identifying vacant ones.

Methods

The core of the proposed method involves using ambient WiFi signals to infer house occupancy. Here's a breakdown of their approach:

  • Opportunistic Smartphone Sensing: Volunteers walk around a community with smartphones equipped with a specific app (Wigle WiFi app) to collect WiFi signal strength (RSSI) data along with GPS coordinates. This data is collected opportunistically, meaning it's gathered passively as people go about their daily routines.
  • Hybrid Network Construction: They construct a "hybrid network" that includes both "real" WiFi access points (RAPs) and "virtual" access points (VAPs). VAPs represent measurement points by a mobile device. This network is built using two types of edges: V2 edges connecting pairs of virtual APs, and VR edges connecting virtual APs to real APs. The weights for VR edges are determined by the RSSI values, using a model that relates signal strength to distance.
  • Statistical Shape Modeling and Localization: The system then employs statistical shape modeling and multidimensional scaling to estimate the locations of these WiFi access points. This involves computing shortest-path distances within the hybrid network and then using techniques like Procrustes analysis to derive the absolute geographical locations of the real APs.
  • Occupancy Inference: The fundamental assumption is that houses with at least one active WiFi access point are likely occupied. Therefore, by successfully localizing occupied houses, the remaining houses can be inferred as vacant.

Datasets

The authors conducted a field trial in a suburban residential community. They collected data using two Android smartphones (Nexus 5 and Xperia Z Ultra) over approximately one hour. This resulted in data for 962 WiFi APs and 610 virtual APs. The ground truth for occupied and vacant houses was provided by local community members.

Key Insights and Ideas

  • WiFi as a Proxy for Occupancy: The paper leverages the widespread adoption of WiFi in households as a strong indicator of occupancy. This is a practical and scalable approach, especially with the increasing prevalence of smart home devices.
  • Hybrid Network for Robust Localization: The creation of a hybrid network of real and virtual APs, combined with statistical modeling, allows for more robust localization of WiFi access points, even with sparse measurements or less accurate GPS data.
  • Reducing Field Survey Burden: The primary insight is that by accurately identifying occupied houses, the number of properties requiring manual inspection for vacancy can be drastically reduced, leading to significant cost and time savings for local authorities.

Key Results

In their field trial, the method successfully identified 57% of occupied houses. Crucially, the method produced no false positives, meaning no vacant houses were incorrectly identified as occupied. Out of 286 houses in the community, 278 were occupied and 8 were vacant. The method detected 57 occupied houses, which, while lower than the total number of detected APs, still represents a substantial reduction in the workload for field surveys.

Limitations and Future Work

  • False Negatives: The paper acknowledges that the detection accuracy is not perfect, with some occupied houses being missed (false negatives). Potential reasons for this include very weak WiFi signal strengths or the absence of a WiFi access point in an occupied house (e.g., if only using cable or mobile connections).
  • Privacy Concerns: The authors recognize that individuals might be wary of their WiFi SSIDs being used for this purpose. They suggest using encrypted hash IDs and allowing users to disable beaconing features as potential solutions.
  • Improving AP Localization: The authors believe that improving the accuracy of WiFi access point localization could further enhance the detection rate and reduce the workload for field surveys.
  • Volunteer Recruitment: A key future direction is to recruit volunteers from local communities to assist with data collection, further scaling the effort.

Summary updated

BibTeX
@inproceedings{konomi2017exploring,
  author = {Shin'ichi Konomi and Tomoyo Sasao and Simo Hosio and Kaoru Sezaki},
  title = {{Exploring the Use of Ambient WiFi Signals to Find Vacant Houses}},
  booktitle = {AmI},
  year = {2017},
  doi = {10.1007/978-3-319-56997-0\_10},
}
Figure