Mobility analytics based on passive sensing data
Loading...
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Alternative Title(s)
Abstract
Nowadays, mobility analytics plays an important role in our daily lives, such as in improving urban planning, optimizing transportation systems, and enhancing public safety. With the rapid development of sensor technology, contemporary mobility analytics based on sensing data has become more efficient and reliable. Depending on how the sensing data is acquired, mobility analytics is generally divided into active sensing-based mobility analytics and passive sensing-based mobility analytics. Active sensing-based mobility analytics, such as deploying Light Detection and Ranging (LiDAR) sensors for traffic flow monitoring or Radio Detection And Ranging (RADAR) systems for movement tracking, either requires a high cost of deploying additional infrastructure or users’ active and continuous participation in sensing data collection. In addition, privacy concerns may arise due to the exposure
of personal identity in active sensing technologies.
Passive sensing-based mobility analytics, such as Wi-Fi-based localization or Inertial Measurement Unit (IMU)-based navigation, passively collects data from existing Internet of Things (IoT) sensors in the environment. Because these IoT sensors are not dedicated to mobility analytics but pre-exist to support other IoT services, passive sensing-based mobility analytics avoids the high costs of installing additional infrastructure, the need of users’ active participation, and the exposure of users’ identities.
However, passive sensing-based mobility analytics still faces many limitations. First, since passive sensing technology lacks dedicated IoT infrastructure, sensing data from a single sensor usually provides mobility information from a limited perspective. Second, the process of collecting passive sensing data is uncontrollable because ambient sensors are not directly controlled by users. As a result, the potential data loss may cause failures in mobility analytics when one type of sensor stops providing data for unknown reasons. Third, passive sensing data is generally sparse also due to this uncontrollable data collection process. Therefore, many studies aim to generate denser sensing data to enhance mobility analytics. However, generating reliable sensing data is non-trivial and remains challenging.
To address these limitations, this dissertation proposes a collaborative and complementary computing paradigm for passive sensing-based mobility analytics. The key idea behind the proposed paradigm lies in three aspects: 1) mobility analytics based on multi-modal sensing data, 2) mobility analytics based on crossdomain sensing data, 3) multi-model-based sensing data generation. In the first case, different types of sensors are jointly utilized for mobility analytics to complement the weaknesses of individual sensors. In the second case, different forms of sensing data from the same sensor are incorporated to provide insights from different knowledge domains. In the third case, Artificial Intelligence (AI)-driven methods and non-AI-driven methods are synergized to generate denser sensing data.
In this dissertation, the feasibility of the proposed computing paradigm is first demonstrated in our preliminary work, where the advantages of collaboration and complement between different sensors are exhibited. Next, this dissertation further
investigates the necessity and effectiveness of the proposed computing paradigm through comprehensive mobility analytics in the following three scenarios.
• This dissertation estimates Physical Proximity between users based on multi-modal sensing data, i.e., Wi-Fi data and IMU data. Wi-Fi data provides
absolute spatial information for mobility analytics, which IMU data lacks. Conversely, IMU data offers fine-grained mobility information, which is not
available in Wi-Fi data. The joint use of Wi-Fi data and IMU data complements each other’s weaknesses, facilitating more reliable physical proximity
estimation.
• This dissertation investigates users’ Visual Attention based on cross-domain sensing data, i.e., eye movements and light patterns reflected in human eyes.
The movement of human eyeballs and the light patterns reflected in human eyes characterize users’ visual attention from different knowledge domains.
Therefore, the joint use of both types of data enables a more comprehensive analysis of users’ visual attention.
• This dissertation develops a framework for Indoor Localization through multi-model-based Wi-Fi fingerprint generation. First, Wi-Fi radio maps
are augmented by jointly utilizing a Generative Adversarial Network (GAN) model and a Gaussian Process Regression (GPR) model,leveraging
the strengths of each approach. Second, a tailored localization algorithm is designed by incorporating the augmented Wi-Fi radio maps.
This dissertation provides a comprehensive and in-depth mobility analytics based on the proposed paradigm. On the one hand, the importance of collaboration and complementarity in passive sensing-based mobility analytics is validated. On the other hand, feasible strategies for mobility analytics in different scenarios are given in this dissertation.
Description
Table of contents
Keywords
Localization, Mobility analtytics, Trajectory similarity, Visual trajectory, Wireless sensing
Subjects based on RSWK
Lokalisation, Mobilität, Trajektorie (Kinematik), Sensortechnik