TL;DR
This paper introduces iLoc, a novel system that uses only smartphone sensors to determine the precise seat location (left/right, front/back) of multiple phones within a vehicle, without requiring any external infrastructure. By analyzing cooperative sensor data during specific driving events like turns and bumps, iLoc achieves high positioning accuracy, paving the way for more context-aware in-vehicle applications.
What Problem Does This Paper Address?
In-vehicle applications, such as personalized entertainment or driver assistance systems, often require knowing where passengers are seated. Existing solutions typically rely on specialized hardware like OBD-II adapters or Bluetooth stereos, which aren't universally available in all vehicles. Manual input from users is inconvenient, and other methods might be infrastructure-heavy or require pre-sampling. This paper tackles the challenge of achieving accurate, infrastructure-free seat-level positioning using only the sensors already present in smartphones, enabling a more accessible and ubiquitous solution.
Methods
iLoc employs a multi-stage approach leveraging smartphone sensors:
- Event-Triggered Sensing: Instead of continuous data collection, iLoc activates its positioning algorithms only when specific events occur. These events include vehicle turns (for left/right identification) and passing over uneven road surfaces (for front/back identification). This significantly reduces energy consumption.
- Reference Frame Transformation: To account for arbitrary phone placement (pockets, bags, etc.), sensor data is transformed from the phone's local coordinate system to a device-independent vehicle frame using accelerometer and gyroscope data.
- Cooperative Positioning: Smartphones within the vehicle exchange sensor data to collaboratively determine their relative positions.
- Left/Right Identification: This is achieved by analyzing the differences in centripetal acceleration experienced by phones during vehicle turns. Phones on the outer side of the turn will experience higher centripetal acceleration.
- Front/Back Identification: This is done by analyzing vertical accelerations when the vehicle encounters uneven surfaces. The timing and characteristics of the vibrations caused by front and rear wheels hitting obstacles differ. A novel "sliding-SD" (Standard Deviation) method is proposed to robustly identify these differences.
- Synchronized Sensing: A critical component is synchronizing the sensor data from multiple phones. This involves time stamp calibration, re-sampling to create equally-spaced data, and a "shift-and-compare" algorithm to align the time series based on cross-correlation.
- Amplitude Calibration: To account for variations in accelerometer sensitivity between different phones, an amplitude calibration step is performed using data collected during translational movement.
Datasets
The paper describes experiments conducted using real driving data. The authors have made their collected data set publicly available for download at http://www.comp.polyu.edu.hk/~cszhe/sensordata/. This dataset includes extensive driving trajectories, turns, and vertical accelerations from city driving scenarios.
Key Insights and Ideas
- Leveraging Everyday Events: The core innovation is using common driving events (turns, bumps) as natural triggers for positioning, eliminating the need for constant sensing and external infrastructure.
- Cooperation is Key: By having phones communicate and share sensor data, iLoc can overcome the limitations of individual phone sensors and achieve more robust positioning.
- Robustness to Noise and Placement: The system is designed to be tolerant of noisy sensor data and arbitrary phone orientations, making it practical for real-world use.
- Energy Efficiency: The event-triggered approach dramatically reduces battery drain, a crucial factor for mobile applications.
- "Sliding-SD" for Front/Back: The proposed sliding-SD method offers a more robust way to distinguish front/back positions compared to simply looking for peak accelerations, as it captures the overall vibration characteristics.
Key Results
- High Accuracy: In the majority of cases, iLoc achieves positioning accuracy of 90%. Even in worst-case scenarios, it reaches around 70%.
- Left/Right Identification: Accuracy is influenced by phone distance and vehicle speed. With sufficient separation and higher speeds, accuracy approaches 100%. The proposed signal processing significantly outperforms raw data comparison.
- Front/Back Identification: Accuracy is affected by vehicle speed and phone placement. Generally high accuracy is reported, with some configurations achieving over 95%.
- Energy Efficiency: With event-triggering, phones maintained over 70% battery after 60 minutes of operation, compared to only about 60 minutes without event-triggering.
Limitations and Future Work
- Reliance on Events: The system's positioning capability is dependent on the occurrence of specific driving events (turns, bumps). Long stretches of straight, smooth highway might pose a challenge.
- Multiple Smartphones Required: iLoc currently requires the cooperation of multiple smartphones within the vehicle.
- User Movement Noise: Phones placed in pockets or held by hand can introduce additional noise from user movements, which the current system doesn't fully account for.
- Future Work:
- Detecting and filtering user movement noise.
- Handling scenarios with no triggering events (e.g., long highway drives) by exploring Ackermann steering geometry during lane changes.
- Developing a mechanism for a single smartphone to self-locate based on a learned model of its sensor data and in-vehicle position.
Summary updated