Intelligent Travel Activity Monitoring: Generalized Distributed Acoustic Sensing Approaches
Ruikang Zhong, Chia-Yen Chiang, Mona Jaber, Rupert De Wilde, Peter Hayward · 2025
Read Intelligent Travel Activity Monitoring: Generalized Distributed Acoustic Sensing Approaches by Ruikang Zhong, Chia-Yen Chiang, Mona Jaber, Rupert De Wilde, Peter Hayward (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2506-10237.txt.
Introduction
Ruikang Zhong and co-authors listen to the ground through fiber optic cables to count walkers and cyclists in this 2025 paper. Cities need travel data to plan greener transport, but cameras invade privacy and wearables miss most people. Buried fibers already sense vibrations from footsteps and tires.
A deep learning system reads those vibrations and sorts the motion patterns. Field trials in two distant places supplied training data. The hard part is that soil and climate change the signal from site to site. Where networks allow, the team trains the sensing nodes together through federated learning, sharing lessons without sharing data. Where a node stands alone, a meta-learning start lets it adapt from a handful of local samples. Tests on telling walking from cycling showed the system travels well between places.
Catalog record
| CREATOR | Ruikang Zhong, Chia-Yen Chiang, Mona Jaber, Rupert De Wilde, Peter Hayward |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 54 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2506-10237
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Intelligent Travel Activity Monitoring: Generalized Distributed Acoustic Sensing Approaches Ruikang Zhong, Chia-Yen Chiang, Mona Jaber, Rupert De Wilde, Peter Hayward Abstract Obtaining data on active travel activities such as walking, jogging, and cycling is important for refining sustainable transportation systems (STS). Effectively monitoring these activities not only requires sensing solutions to have a joint feature of being accurate, economical, and privacy-preserving, but also enough generalizability to adapt to different climate environments and deployment conditions. In order to provide a generalized sensing solution, a deep learning (DL)-enhanced distributed acoustic sensing (DAS) system for monitoring active travel activities is proposed. By leveraging the ambient vibrations captured by DAS, this scheme infers motion patterns without relying on image-based or wearable devices, thereby addressing privacy concerns. We conduct real-world experiments in two geographically distinct locations and collect comprehensive datasets to evaluate the performance of the proposed system. To address the generalization challenges posed by heterogeneous deployment environments, we propose two solutions according to network availability: 1) an Internet-of-Things (IoT) scheme based on federated learning (FL) is proposed, and it enables geographically different DAS nodes to be trained collaboratively to improve generalizability; 2) an off-line initialization approach enabled by meta-learning is proposed to develop high-generality initialization for DL models and to enable rapid model fine-tuning with limited data samples, facilitating generalization at newly established or isolated DAS nodes. Experimental results of the walking and cycling classification problem demonstrate the performance and generalizability of the proposed DL-enhanced DAS system, paving the way for practical, large-scale DAS monitoring of active travel. I Introduction Active travel modes, such as walking, jogging, cycling, and skateboarding, play a crucial role in sustainable transportation systems (STS), which not only reduce the congestion caused by vehicle transportation but also benefit public health [ 1 , 2 ] . Accurately gathering the information of active travel activities is essential for transportation management, urban planning, and the development of a smart city [ 3 ] . The utilization of Internet-of-Things (IoT) devices and sensing technology to collect travel information has emerged as a promising solution, demonstrating superior efficiency and accuracy in comparison to conventional questionnaires or manual statistics. However, the prevalent image-based methods or invasive devices often compromise individual privacy [ 4 ] , limiting their application scenarios and public acceptance in practical applications. Consequently, there is an increasing demand for efficient and privacy-preserving sensing technologies for travel activity detection in transportation systems. Fortunately, the distributed acoustic sensing (DAS) system integrated with advanced artificial intelligence (AI) techniques offers a promising solution for active travel monitoring. Overall, in this scheme, the DAS sensors can collect acoustic information caused by the activity, while the deep learning (DL) algorithm is responsible for the analysis of collected signals [ 5 ] . Specifically, the DAS technology re-utilizes existing fiber-optic cables to detect human activities without recording personal biometric data (e.g., appearance), thus inherently addressing privacy concerns [ 6 ] . Furthermore, the DAS system has a coverage area that encompasses tens or even hundreds of kilometers, and in the time domain, its monitoring is long-term and uninterrupted surveillance [ 7 ] . Once the DAS sensors are integrated with well-trained DL models, DAS enables fine-grained interpretation of complex activity patterns, demonstrating significant superiority in terms of coverage, accuracy, privacy, and efficiency for monitoring active travel motions [ 8 ] . Despite the promise of AI-enhanced DAS systems, several practical challenges remain unresolved due to the variability of deployment environments at different DAS nodes in large-scale DAS implementation. Since optical fibers are buried underground, each DAS implementation is influenced not only by the status of the fiber itself but also by environmental factors, including but not limited to the road conditions, temperature, humidity, and soil hardness [ 9 ] . Such heterogeneity undermines the generalization capability of DL models that are well-fitted to any specific environment. Although it is a feasible solution to perform independent local training for each DL model employed by different DAS nodes, the hardware expense, time cost, and computational complexity required for collecting the local dataset and conducting complete training are enormous and unaffordable. Therefore, addressing the challenge of generalizability in AI-enhanced DAS systems is a prerequisite for the widespread deployment of the system. To address the generalization issue in DAS practice, in this paper, we propose two approaches based on federated learning (FL) and meta-learning, which fit different scenarios depending on the network availability. By enabling models to collaboratively learn from decentralized data or training a generalized initialization that can quickly adapt to new environments, the proposed schemes ensure the accuracy of active travel activity classification across diverse DAS deployments at the same accurate level of local data training. Fig. 1 : Model and working principle of the DAS system. I-A Related Research I-A 1 DAS enabled ITS The DAS system boasts extensive coverage and economic advantages, making it a successful solution for a variety of applications, such as pipeline monitoring [ 10 ] , earthquake detection [ 11 ] , and perimeter detection [ 12 ] . In the realm of transportation, a series of studies have been formulated with the objective of employing DAS systems to monitor the operation of vehicles [ 13 ] and trains [ 14 ] . Vehicle classification is a classic application of DAS in transportation systems. A vehicle detection and classification algorithm based on signal processing and feature extraction was proposed in [ 15 ] , with an accuracy of 80% and 70%, respectively. In [ 16 ] , a wavelet threshold algorithm was proposed to support DAS systems in estimating the speed of multiple vehicles. Furthermore, the authors of [ 17 ] proposed an image processing scheme to estimate the traffic flow and average speed of vehicles using image representations generated from segments of DAS signals. The emergence of machine learning technologies further assisted the DAS system, and integrating DAS and data-driven machine learning methods exhibits potential to develop AI-enhanced traffic perception systems [ 18 ] . Deep learning-based algorithms such as convolutional neural networks [ 19 ] , recurrent neural networks, and transformers [ 20 ] are helpful to extract spatiotemporal features from DAS signals to achieve tasks such as vehicle type discrimination, pedestrian behavior recognition, and road anomaly detection [ 21 ] . A U-Net image segmentation model was proposed to detect low-speed quasi-static vehicles [ 22 ] . In our previous work [ 5 ] , a sparse residual network (SR-Net) model and an Alex-SR model were proposed for the vehicle classification and occupancy detection problem, respectively. This approach solves the problem that the low strain of quasi-static vehicles is easily masked by environmental noise and DAS fading noise. However, it is worth noting that the current applications of DAS focus on heavy equipment, while DAS detection is more challenging for active travelers who have significantly lower weight. I-A 2 Generalization of Models The effects of ambient temperature, road material, deployment depth, and fiber usage time on the received signal-to-noise ratio of the DAS system are revealed in [ 13 ] . The authors found that the above factors can have an impact of more than 3dB on the received signal-to-noise ratio through vehicle experiments, which supports the necessity of improving the generalization