PaperlayerView exact revision on arXiv

Source-linked AI summary

SARDO: An Automated Search-and-Rescue Drone-based Solution for Victims Localization

Antonio Albanese, Vincenzo Sciancalepore, Xavier Costa-Pérez

arXiv:2003.05819v1cs.NI

TL;DR

SARDO addresses victim localization in large or difficult disaster areas when cellular infrastructure and conventional location information are unavailable. It uses an autonomous UAV-based mobile network with pseudo-trilateration and machine learning to localize phones without modifying them. Field trials report localization in a few minutes per phone, accuracy of a few tens of meters, and approximately 5% battery cost.

  • Problem

    Victim localization is difficult when people are sparsely distributed, cellular networks are down, GNSS is unavailable, or victims cannot transmit their locations.

  • Method

    SARDO uses a single UAV as an infrastructure-free cellular platform, combining pseudo-trilateration with machine-learning prediction and feedback-controlled movement.

  • Results

    SARDO localizes phones in a few minutes with accuracy of a few tens of meters and approximately 5% battery cost at the reported operating point.

  • Takeaways & Limitations

    SARDO provides an automated drone-based cellular localization approach for disaster areas without functioning mobile infrastructure or available UE GNSS information.

  • Takeaways & Limitations

    Localization accuracy may decrease for mobile targets, while fading, shadowing, and time variability can make distance measurements and trilateration inconsistent.

Abstract

from arXiv · show

Natural disasters affect millions of people every year. Finding missing persons in the shortest possible time is of crucial importance to reduce the death toll. This task is especially challenging when victims are sparsely distributed in large and/or difficult-to-reach areas and cellular networks are down. In this paper we present SARDO, a drone-based search and rescue solution that exploits the high penetration rate of mobile phones in the society to localize missing people. SARDO is an autonomous, all-in-one drone-based mobile network solution that does not require infrastructure support or mobile phones modifications. It builds on novel concepts such as pseudo-trilateration combined with machine-learning techniques to efficiently locate mobile phones in a given area. Our results, with a prototype implementation in a field-trial, show that SARDO rapidly determines the location of mobile phones (~3 min/UE) in a given area with an accuracy of few tens of meters and at a low battery consumption cost (~5%). State-of-the-art localization solutions for disaster scenarios rely either on mobile infrastructure support or exploit onboard cameras for human/computer vision, IR, thermal-based localization. To the best of our knowledge, SARDO is the first drone-based cellular search-and-rescue solution able to accurately localize missing victims through mobile phones.

1 INTRODUCTION

SARDO addresses rapid victim localization when communication networks are unavailable and conventional location information cannot be obtained. It uses UAVs to detect mobile devices and combines machine learning with localization measurements in an integrated rescue system.

  • SARDO targets rapid victim localization when terrestrial networks are down, GNSS is unavailable, or victims cannot transmit their locations.
  • UAVs can be rapidly deployed, reach difficult areas, cover broad searches, and detect victims’ widely available smart devices.
  • SARDO integrates IMSI-based user detection, pseudo-trilateration, neural position prediction, and trajectory control in one localization system.
  • Field testing with commercial off-the-shelf components provides feasibility evidence and quantitative system-performance measurements.

2 SARDO FRAMEWORK OVERVIEW

SARDO uses a single UAV as a portable cellular base station to collect time-of-flight measurements and estimate mobile-phone positions. Machine-learning prediction and feedback-controlled relocation adapt the search trajectory during sequential localization tasks.

  • A single UAV sweeps a predefined area as a portable cellular base station, replacing multi-anchor measurements with information collected from one moving platform.
  • Commercial-phone compatibility is preserved because SARDO introduces no protocol-stack modifications while measuring uplink time of flight.
  • SARDO combines time-diverse measurements with CNN and feed-forward networks to estimate current user positions and motion trajectories.
  • An LSTM forecasts future user positions from previous estimates, enabling the system to anticipate target movement.
  • Feedback control adjusts the UAV trajectory by narrowing the scan after accurate forecasts or enlarging it after prediction errors.
  • Localization runs as sequential atomic tasks within short thresholds, with each task’s duration affecting measurement count and localization accuracy.

3 GEO-LOCALIZATION MODEL

The section contrasts classical multi-lateration with SARDO’s pseudo-trilateration, which uses one moving anchor to localize noisy and mobile targets. It then characterizes the resulting solution ambiguities and motivates an AI-based extension for learning motion and improving localization.

  • 3.1 Legacy multi-lateration: Classical multi-lateration requires multiple known anchors and becomes vulnerable to channel noise, fading, shadowing, and target mobility.With noisy distances, sphere intersections may be ambiguous; with moving targets, measurements collected across space or time can become inconsistent.
  • 3.2 A novel technique: Pseudo-Trilateration: Pseudo-trilateration uses one anchor that collects distance measurements over time while moving along a trajectory, avoiding highly dense anchor coverage and accommodating mobile users.The approach combines measurements from different anchor positions rather than simultaneous measurements from multiple anchors.
  • 3.2 A novel technique: Pseudo-Trilateration: The pseudo-trilateration optimization problem is NP-Hard, motivating the use of computational optimization and learned localization components.The paper states that the problem can be formulated as a convex optimization model, while separately proving its NP-hardness.
  • 3.2 A novel technique: Pseudo-Trilateration: For a static target and linear anchor motion, pseudo-trilateration yields two distinct solutions, except when the target lies on the anchor trajectory, where one double solution remains.The same directional condition produces the exact target position with double multiplicity in the illustrated case.
  • 3.2 A novel technique: Pseudo-Trilateration: SARDO extends pseudo-trilateration with AI that learns common motion behaviors and estimates target positions from measurements collected over a time window.The paper describes combining positions covered by a moving target with an AI-based localization system intended to calculate positions with high accuracy.

4 AI-BASED ENHANCED LOCALIZATION

SARDO combines UAV-based ranging with CNN and LSTM models to estimate current user trajectories and predict future positions for enhanced localization.

  • 4 AI-BASED ENHANCED LOCALIZATION: A single UAV collects Time of Flight measurements along a predefined trajectory to localize a target user under cellular coverage.The approach assumes fixed UAV altitude, while altitude can be incorporated into the retrieved information.
  • 4.1 Data preprocessing: The ranging series is arranged into N × L matrices and concatenated with UAV measurement-spot coordinates to preserve temporal and spatial relationships.The resulting representation combines the measurement matrix Γm with coordinate matrix Dm.
  • 4.2 2D CNN design and training: A 2D CNN regresses the user trajectory from processed ranging data, using simulated UAV and user trajectories to create labeled training samples.Obstacle-induced shadowing produces stripe patterns in Γm, motivating treatment of the inputs as single-channel images.
  • 4.3 Encoder-Decoder LSTM design and training: An Encoder-Decoder LSTM maps the estimated current user trajectory to a sequence of predicted future user positions.The LSTM encoder produces a fixed-length representation that the decoder maps onto the target sequence.

5 DYNAMIC UAV RELOCATION

SARDO dynamically relocates the UAV using predicted user positions, adapting its circular trajectory to improve subsequent distance measurements.

  • 5 DYNAMIC UAV RELOCATION: The controller changes the circular trajectory radius and center offsets after each revolution to bring the UAV closer to the user.This adjustment is intended to retrieve more accurate distance measurements.
  • 5 DYNAMIC UAV RELOCATION: Predicted user positions are averaged to set the next UAV trajectory center, while the maximum distance from that center determines its radius.The controller also uses relative user and UAV speeds to add a safety margin when the user moves faster.
  • 5 DYNAMIC UAV RELOCATION: For users moving within a limited area, the UAV reduces the circular trajectory area by decreasing δρ to focus measurements around the expected positions.The controller is described as adaptively tuning parameters across multiple UAV positions.

6 SARDO IMPLEMENTATION

SARDO integrates cellular identification, time-of-flight measurement, machine-learning localization, and autonomous UAV control into an infrastructure-independent prototype. Its implementation uses off-the-shelf hardware and does not require protocol-stack changes or successful UE attachment.

  • 6 SARDO IMPLEMENTATION: The prototype combines an IMSI-catcher, software-defined radio, onboard processing, and UAV control components.The architecture uses srsLTE on an Intel NUC, an NI USRP B210, and a DJI Matrice 600 Pro UAV.
  • 6 SARDO IMPLEMENTATION: A PI controller adjusts the UAV trajectory to reduce steady-state positioning error, while runtime applications can automatically trigger new directions.The controller is selected for simplicity and zero steady-state error.
  • 6.1 Mobile Infrastructure Independence: SARDO operates without successful UE attachment, reducing system complexity for emergency rescue operations.The UAV identifies UEs through LTE signaling even though the attachment procedure is rejected.
  • 6.1 Mobile Infrastructure Independence: SARDO triggers UE identity disclosure through LTE tracking-area-update signaling while remaining compliant with 3GPP guidelines.The UAV changes the tracking area and responds with a rejection that prompts the UE to reveal its IMSI.
  • 6.2 Tunable precision: The system estimates uplink time of flight from DMRS cross-correlation, with 20-MHz LTE sampling yielding approximately 9.8 m distance resolution before upsampling.The method locates the cross-correlation magnitude peak and can upsample signals to improve resolution.

7 PERFORMANCE EVALUATION

SARDO is evaluated through synthetic simulations and a rural field trial across channel conditions, user trajectories, speeds, and UAV altitudes. The results show tens-of-meters localization, robustness to rubble-induced error, and approximately 5% battery impact at the reported 100 m operating point.

  • 7 PERFORMANCE EVALUATION: The evaluation combines synthetic traces with a prototype field trial and measures localization error and trajectory-reconstruction similarity.Simulation settings vary user speed, channel conditions, mobility patterns, and UAV trajectories.
  • 7.2 Pseudo-trilateration validation: 31 m median localization accuracy remains stable without rubble, while rubble increases error and variance but leaves the similarity index close to 1.The CNN was trained without rubble; its trajectory reconstruction remains robust despite scattering and attenuation.
  • 7.3 SARDO stability performance: Localization error increases with user speed, while SARDO can reduce trajectory radius and improve accuracy for both static and nomadic users.The UAV speed is set to the maximum user speed supporting reasonably accurate localization, and higher UAV speeds are possible.
  • 7.4 Proof of Concept Experimental Results: Distance-measurement variance increases with actual distance, and altitude affects performance through a trade-off between antenna illumination and received power.The field dataset contains measurements from two victim locations, while altitude experiments assess CNN performance across speeds.
  • 7.4 Proof of Concept Experimental Results: Approximately 5% relative battery impact is measured at the 100 m operating point.Battery impact decreases as drone altitude increases, making the reported 100 m point the stated operational optimum.

8 RELATED WORK

Prior localization work includes mathematical, AI-based, and UAV-placement approaches, while commercial disaster solutions commonly depend on infrastructure or onboard sensing. SARDO is positioned as a single-UAV cellular system operating without mobile infrastructure or GNSS support.

  • 8 RELATED WORK: Related work spans closed-form multilateration, neural-network localization, UAV deployment optimization, and high-accuracy indoor UWB systems.These approaches address positioning accuracy, harsh environments, flight constraints, or indoor first-responder localization.
  • LTE and Drone-based Commercial Localization Solutions: Existing disaster-localization solutions generally assume mobile infrastructure or use drone cameras, infrared, or thermal sensing.The paper contrasts these approaches with SARDO’s cellular localization design.
  • LTE and Drone-based Commercial Localization Solutions: The paper identifies no prior drone-based cellular localization solution designed to operate without mobile infrastructure or GNSS support.This is presented as the paper’s novelty claim based on its related-work review.

9 CONCLUSIONS

SARDO is a drone-based cellular localization system designed for disaster areas where infrastructure or GNSS information may be unavailable. A field trial supports its feasibility, with localization accuracy of a few tens of meters, few minutes per phone, and low battery cost.

  • 9 CONCLUSIONS: A few tens of meters accuracy, few minutes per phone, and low battery cost characterize SARDO’s reported localization performance.The system was tested in a field trial using commercial off-the-shelf components.
  • 9 CONCLUSIONS: SARDO provides drone-based cellular localization when mobile infrastructure is out of service and user-equipment GNSS information is unavailable.It is intended for disaster scenarios with sparse victims in large or difficult-to-reach areas.
  • 9 CONCLUSIONS: The system supports multiple-victim localization by running sequentially within a defined search area and automates operations using GNSS-defined boundaries.Its machine-learning feedback loop improves localization accuracy while controlling the UAV trajectory.
  • 9 CONCLUSIONS: A rural field trial with commercial off-the-shelf components demonstrated SARDO’s feasibility and produced quantitative evidence about practical performance.
Loading 2003.05819v1…