TL;DR
This paper presents a low-cost wearable system for pedestrian tracking that integrates inexpensive Inertial Measurement Units (IMUs) with ultrasound sensors. By combining these with an Extended Kalman Filter (EKF) and an optional particle filter for map-based constraints, the system achieves significant improvements in accuracy, reducing accumulated errors by up to 15% and keeping tracking errors below 2% after initial steps. This work demonstrates the viability of affordable sensors for robust pedestrian navigation.
What Problem Does This Paper Address?
Accurate pedestrian navigation is crucial for many applications, from healthcare monitoring to assisted navigation and emergency response. While high-end Inertial Measurement Units (IMUs) can provide good tracking, their prohibitive cost limits widespread adoption. Existing low-cost IMUs suffer from significant drift over time, leading to accumulated positional errors. This paper aims to address this by investigating the potential of using a combination of low-cost IMUs and ultrasound sensors to achieve accurate pedestrian tracking without relying on expensive equipment. The challenge lies in effectively fusing data from these disparate sensors and mitigating the inherent drift of IMUs.
Methods
The proposed system, dubbed the "IMU/US system," is built around several key components:
- Low-Cost IMUs: The system utilizes inexpensive ADXL345 accelerometers and ITG-3200 gyroscopes, significantly reducing hardware costs.
- Equipment-Free IMU Calibration: A novel calibration method is employed that doesn't require specialized mechanical platforms. It leverages the constant magnitude of gravity and gyroscope-induced rotations to estimate scale factors, misalignments, and biases.
- Extended Kalman Filter (EKF): This forms the core of the navigation system. It incorporates:
- Zero-Velocity Updates (ZUPTs): Assumes zero velocity when the foot is on the ground to correct for drift.
- Heuristic Drift Reduction (HDR): Limits yaw drift by assuming small changes in yaw between steps are due to error.
- Ultrasound Sensors: HC-SR04 ultrasound sensors are mounted on the feet to measure inter-foot distances. When both feet are on the ground, these range measurements are used to estimate relative foot positions, providing corrections to the IMU-based tracking. A linear system and non-linear least-squares algorithm are used to process these range measurements.
- Particle Filter (Optional): For environments with known maps, a particle filter is employed. It uses map information to constrain the possible user motions, significantly improving accuracy and preventing unbounded drift. This is achieved by assigning weights to particles based on whether they intersect walls.
Datasets
The paper evaluates the system's performance using two types of walking scenarios:
- Type 1 Walking: Conducted in a gait laboratory with a rectangular area (approximately 4m x 2m). Ground truth data was provided by a Vicon motion capture system.
- Type 2 Walking: Performed in a more natural indoor environment (approximately 55m walk through rooms and corridors). In this case, ground truth was not available, and performance was assessed based on loop misclosure.
The authors do not mention using pre-existing datasets for training or evaluation, but rather collected their own data through controlled experiments.
Key Insights and Ideas
- Low-Cost Sensors Can Be Effective: The study strongly demonstrates that affordable IMUs and ultrasound sensors, when integrated intelligently, can achieve performance comparable to more expensive systems. This opens doors for more accessible pedestrian navigation solutions.
- Ultrasound for Step Length Correction: The integration of ultrasound sensors provides a tangible improvement (around 15% reduction in total error) by offering more accurate step-length data, especially when direct line-of-sight is available.
- Particle Filters for Map-Based Constraints: The particle filter's ability to leverage environmental maps is a powerful technique for correcting drift, particularly yaw drift, and keeping tracking errors consistently low (below 2% after initial steps).
- Closed-Loop Walking Masks Errors: The authors highlight that evaluating systems solely on loop misclosure in closed-loop walking can be misleading, as systematic errors are often cancelled out. This suggests that a more comprehensive error assessment, like the "maximum error" metric used, is necessary.
- Yaw Drift Remains a Challenge: While ultrasound helps with step length, it doesn't directly correct yaw drift. This remains an area for future improvement.
Key Results
- IMU Only: Showed loop misclosure errors of around 1% and maximum errors of 4-5% during walking.
- IMU + Ultrasound (IMU/US): Achieved a ~15% reduction in total accumulated error compared to the IMU-only system.
- IMU + Ultrasound + Particle Filter (IUP): Significantly improved accuracy, keeping tracking errors below 2% after the first few steps in Type 1 walking. Cumulative absolute errors were reduced by more than half in some cases (e.g., from ~14.8m to ~6.9m for IMU vs. IUP in Trial 1).
- Type 2 Walking: The IMU/US system performed similarly to the IMU system due to frequent line-of-sight issues with ultrasound in "curvy" environments. Loop misclosure errors were comparable to other studies using high-end IMUs.
Limitations and Future Work
- Ultrasound Line-of-Sight: The effectiveness of the ultrasound system is dependent on a clear line of sight between sensors, which can be an issue in complex or "curvy" environments.
- Yaw Drift: While the particle filter helps maintain yaw accuracy, the paper suggests that incorporating magnetometers could further improve yaw drift correction.
- Form Factor: The current system's housing is 3D printed and may not be socially acceptable for daily use. A slimmer, more aesthetically pleasing design is needed for a viable wearable product.
- Real-Time Particle Filter: The current implementation of the particle filter is too slow for real-time operation and requires optimization or adaptive particle management.
- Complex Environments: Further validation in more complex environments and with longer walks is recommended.
- Commercialization: The authors note that the cost could be further reduced through commercial manufacturing processes.
Summary updated