TL;DR
This paper tackles the challenge of accurate pedestrian localization using everyday smartphones, which suffer from unreliable GNSS signals due to multipath interference and drifting MEMS sensors. They introduce a novel multi-phase fusion framework that first mitigates GNSS multipath errors using a new estimation technique and then fuses the corrected GNSS data with Pedestrian Dead Reckoning (PDR) by jointly estimating stride length and heading to combat drift.
What Problem Does This Paper Address?
Precise pedestrian positioning is crucial for Location-Based Services (LBS), but relying on smartphone sensors presents significant hurdles. Mass-market GNSS modules, found in most smartphones, are prone to multipath interference in urban environments, leading to inaccurate absolute positioning (often 3-10 meters off). Simultaneously, smartphone MEMS sensors (accelerometer, gyroscope, magnetometer) used for Pedestrian Dead Reckoning (PDR) suffer from long-term drift, causing accumulated errors that render the PDR system unreliable over time. Neither system alone can provide satisfactory accuracy, necessitating a robust fusion approach.
Methods
The authors propose a three-phase fusion framework:
- Real-Time Difference (RTD) GNSS: Instead of relying on single-point GNSS, they employ a short-baseline pseudorange double-difference technique using a reference station (CORS). This significantly reduces common errors like ionospheric and tropospheric delays.
- Multipath Mitigation: This is a core innovation. They introduce a novel method combining:
- Multipath Partial Parameters Estimation (MPPE): This technique uses the Carrier-to-Noise-Density Ratio (CN0R) to estimate multipath parameters (amplitude ratio and phase shift).
- Double-Difference Code-Minus-Carrier (DDCMC) Filter: This filter is designed to estimate the remaining multipath error components, specifically the un-difference pseudorange multipath error.
- Joint Stride Length and Heading Estimation: Before the final fusion, they implement joint estimation modules for stride length and heading. This approach leverages the strengths of both GNSS and PDR to suppress long-term drift and noise, smoothing the outputs before they enter the main fusion filter.
The final fusion is performed using a Kalman Filter (KF).
Datasets
The paper describes experimental setups using a Xiaomi 8 UD smartphone and a Huace B5 survey-grade GNSS receiver as a reference station. They also mention using RTK measurements for ground truth. The experiments involve:
- Control Point Test: A rectangular track with known corner points to analyze standalone PDR and RTD errors.
- Static Multipath Test: Static measurements to evaluate the effectiveness of the DDCMC filter in mitigating multipath errors.
- GNSS/PDR Fusion Test: A 400m track to compare the performance of their proposed method against other fusion techniques and standalone systems.
No specific public datasets are mentioned; the experiments are conducted with self-collected data.
Key Insights and Ideas
- Smartphone GNSS limitations are severe: The paper clearly highlights how multipath interference and the low-cost nature of smartphone GNSS hardware significantly degrade positioning accuracy.
- PDR drift is a major challenge: The cumulative nature of heading drift in PDR is identified as a primary reason for trajectory distortion over time.
- A multi-pronged approach is necessary: No single sensor or simple fusion technique is sufficient. The paper advocates for a phased approach that addresses specific error sources sequentially.
- CN0R is a valuable signal for multipath estimation: The MPPE method demonstrates that CN0R, often overlooked for direct positioning, can be effectively used to infer multipath characteristics.
- Joint estimation before fusion is key for drift suppression: By explicitly estimating stride length and heading together before the final fusion, the system can better mitigate the inherent drift of PDR.
Key Results
The proposed multi-phase fusion method demonstrates significant improvements:
- Multipath Mitigation: The MPPE and DDCMC filter effectively suppress double-difference multipath errors exceeding 4 meters.
- Positioning Accuracy: The proposed method achieves a minimum Root Mean Square Error (RMSE) of 1.63 m in positioning accuracy.
- Long-Term Robustness: For 20 minutes of continuous walking, the method achieves a minimum RMSE of 4.71 m, indicating good performance over extended periods.
- Comparative Performance: In tests involving five different methods (Only PDR, Only RTD, MREKF, ESC, and the proposed method), the proposed approach consistently outperformed others, especially in longer-duration tests where PDR drift became more pronounced. For instance, in the third lap of the fusion test, the proposed method's RMSE was 4.71m, while MREKF was 4.81m, ESC was 8.21m, and Only RTD was 5.21m.
Limitations and Future Work
- Smartphone Sensor Variability: The paper acknowledges that the performance of MEMS sensors can vary between different smartphone models, which might affect the generalization of the proposed method.
- Integer Ambiguity Fix: While RTD is used, the paper notes that fixing integer ambiguities for carrier-phase GNSS on smartphones is still challenging and time-consuming, making direct RTK impractical. The proposed RTD with multipath mitigation is presented as a more practical alternative for current low-cost hardware.
- Future Work: The authors implicitly suggest that further improvements could come from exploring more advanced Kalman filter designs or incorporating additional sensor data if available. The focus on mitigating specific error sources (multipath and drift) suggests that future work might explore more sophisticated models for these phenomena.
Summary updated