[Paper Review] Dejavu: An Accurate Energy-Efficient Outdoor Localization System

H. H. Aly, Moustafa Youssef|arXiv (Cornell University)|Oct 9, 2013

Indoor and Outdoor Localization Technologies24 references4 citations

TL;DR

Dejavu presents an energy-efficient outdoor localization system for car navigation that uses inertial sensors (accelerometer, compass, gyroscope) for dead-reckoning and leverages unique sensor signatures from physical landmarks—such as tunnels, bridges, and potholes—and virtual landmarks (e.g., distinct cellular signal patterns)—detected via crowd-sourced data to reset error accumulation. It achieves 8.4 m median error in cities and 16.6 m on highways, outperforming GPS in accuracy and extending battery life by 347%.

ABSTRACT

We present Dejavu, a system that uses standard cell-phone sensors to provide accurate and energy-efficient outdoor localization suitable for car navigation. Our analysis shows that different road landmarks have a unique signature on cell-phone sensors; For example, going inside tunnels, moving over bumps, going up a bridge, and even potholes all affect the inertial sensors on the phone in a unique pattern. Dejavu employs a dead-reckoning localization approach and leverages these road landmarks, among other automatically discovered abundant virtual landmarks, to reset the accumulated error and achieve accurate localization. To maintain a low energy profile, Dejavu uses only energy-efficient sensors or sensors that are already running for other purposes. We present the design of Dejavu and how it leverages crowd-sourcing to automatically learn virtual landmarks and their locations. Our evaluation results from implementation on different android devices in both city and highway driving show that Dejavu can localize cell phones to within 8.4m median error in city roads and 16.6m on highways. Moreover, compared to GPS and other state-of-the-art systems, Dejavu can extend the battery lifetime by 347%, achieving even better localization results than GPS in the more challenging in-city driving conditions.

Motivation & Objective

• To address the high energy consumption of GPS in outdoor car navigation, especially in urban and tunnel environments where satellite signals are weak.
• To reduce error accumulation in dead-reckoning localization using inertial sensors by identifying and leveraging unique environmental landmarks.
• To develop a system that uses only low-energy or already-active sensors to maintain energy efficiency.
• To automatically discover and enrich a database of physical and virtual landmarks through a crowd-sourcing approach.
• To achieve localization accuracy surpassing GPS in challenging urban environments while drastically improving battery efficiency.

Proposed method

• Dejavu employs a dead-reckoning approach using accelerometer, compass, and gyroscope data to estimate continuous position based on displacement and direction.
• It detects physical anchors—such as tunnels, bridges, turns, and bumps—by identifying unique sensor signatures via a finite state machine model.
• Virtual anchors are discovered using unsupervised learning on sensor traces (e.g., unique cellular signal strength patterns) collected from crowd-sourced devices.
• The system uses a unified anchor database to reset dead-reckoning error whenever a known landmark is detected, improving localization accuracy.
• Energy efficiency is maintained by relying only on sensors already active for other purposes (e.g., GSM, opportunistic WiFi) or low-power inertial sensors.
• A crowd-sourcing framework enables automatic, transparent expansion of the anchor database over time without user intervention.

Experimental results

Research questions

• RQ1 Can a system achieve sub-10-meter outdoor localization accuracy using only standard smartphone sensors and avoid reliance on high-energy GPS?
• RQ2 Can unique sensor signatures from environmental features (e.g., potholes, bridges, tunnels) be reliably detected and used to reset dead-reckoning error?
• RQ3 Can virtual landmarks—defined by unique sensor patterns such as cellular signal variations—be automatically discovered and used to improve localization?
• RQ4 To what extent can a system reduce energy consumption compared to GPS while maintaining or improving localization accuracy?
• RQ5 Can the system outperform GPS in urban environments where satellite signals are frequently obstructed?

Key findings

• Dejavu achieves a median localization error of 8.4 meters in urban driving environments, which is 42.9% better than GPS under the same conditions.
• On highways, Dejavu achieves a median error of 16.6 meters, demonstrating robustness in open environments.
• The system extends battery life by 347% compared to continuous GPS usage, significantly improving energy efficiency.
• Dejavu outperforms GPS in accuracy in city driving, where satellite signal degradation and multipath effects typically reduce GPS performance.
• The system successfully detects and leverages both physical anchors (e.g., tunnels, bridges) and virtual anchors (e.g., unique cellular signal patterns) to reset error.
• The crowd-sourced, unsupervised learning approach enables automatic and transparent growth of the anchor database over time.