TL;DR
This paper introduces a novel, low-cost instrumented shoe system using 174 pressure sensors to accurately detect gait phases. By employing a unique "zero potential" method to mitigate sensor crosstalk and a custom peak-heuristic algorithm, the system achieves high accuracy in identifying gait events and sub-phases, even during turning, making it promising for home healthcare and rehabilitation.
What Problem Does This Paper Address?
Accurate gait phase detection is crucial for clinical gait analysis, aiding in the diagnosis and rehabilitation of movement disorders. Traditional methods, like force plates and motion capture systems, are expensive and cumbersome, limiting their widespread use. While instrumented shoes offer a more portable solution, existing systems often face challenges:
- IMU-based systems: Susceptible to sensor placement errors and affected by gait speed and direction.
- Force sensor-based systems: Can have limited durability and accuracy issues due to the number and placement of discrete sensors.
- Plantar pressure systems: Often use a limited number of sensors, leading to low resolution and reliance on computational models for estimation. A significant challenge with dense sensor arrays is the "crosstalk effect," where signals from adjacent sensors interfere with each other.
This paper aims to overcome these limitations by developing a high-resolution, low-cost, and portable instrumented shoe system that accurately detects gait phases, including detailed sub-phases, even in complex walking scenarios.
Methods
The core of this work lies in two main innovations: the hardware design of the instrumented shoe and the gait phase detection algorithm.
Hardware Design:
- High-Resolution Sensing: The shoe insole is embedded with 174 independent sensing units. These units are constructed from an off-the-shelf force-sensitive film (Velostat conductive copolymer) sandwiched between two layers of conductive tape arranged in a matrix. This matrix design, with shared row and column lines, significantly reduces the number of required inter-connection lines and I/O pins.
- Crosstalk Mitigation: To address the inherent "crosstalk effect" in matrix-based sensor arrays, the researchers implemented a "zero potential" method. This involves using single-pole-double-throw (SPDT) switches for rows and single-pole-single-throw (SPST) switches for columns. By selectively connecting a sensing unit to a load resistor and the negative input of an operational amplifier while grounding others, the system can isolate and measure the resistance of individual sensing units without interference from neighboring ones.
- Low-Cost and Portable: The entire system, including the control board and battery, is designed to be low-cost (approximately $30 USD) and portable, weighing similar to a normal sports shoe.
Gait Phase Detection Algorithm:
- Peak Heuristic Search: A customized algorithm was developed based on the foot plantar pressure data. Instead of relying on single sensor outputs, which can be inconsistent, the algorithm averages sensor outputs from four main plantar sub-areas: heel, foot arch, forefoot, and toe. This averaging minimizes the impact of individual sensor variations.
- Key Event Detection: The algorithm identifies key gait events (Heel Contact - HC, Forefoot Contact - FC, Heel Off - HO, Toe Off - TO) by analyzing the peaks and local minima in the averaged sensor outputs from the defined sub-areas. For instance, HC is identified as the local minimum before the first peak in the heel sensor output.
- Sub-Phase Calculation: Once the key events are detected, the algorithm calculates the durations of stance sub-phases: Initial Contact (IC), Flat Foot (FF), and Push Off (PO). These durations are defined as the time intervals between the identified key gait events. The algorithm is designed to be computationally efficient, avoiding the need for extensive training data required by machine learning methods.
Datasets
The study did not introduce a new dataset for training. Instead, it used data collected from experiments with human participants for evaluation.
- Experimental Data: Data was collected from eight healthy male participants performing straight-line walking and turning walking on a 10m platform.
- Reference Measurement: Ground reaction forces were measured using two synchronized force plates (Bertec, Sweden) placed underneath the walking platform. These force plate measurements served as the ground truth for evaluating the accuracy of the instrumented shoe system.
Key Insights and Ideas
- Crosstalk is a Solvable Problem: The "zero potential" method effectively demonstrates a scalable and practical solution for mitigating crosstalk in dense matrix-based pressure sensor arrays, a common challenge in such systems.
- Averaging for Robustness: Using averaged sensor outputs from plantar sub-areas significantly improves the robustness of gait phase detection, making it less sensitive to individual sensor variations and manufacturing inconsistencies.
- Turning Gait is Feasible: The system's ability to accurately detect gait phases during turning walking, a more complex scenario than straight-line walking, highlights its versatility and potential for real-world applications where gait direction is not constant.
- Low-Cost, High-Resolution is Achievable: The combination of off-the-shelf materials (Velostat) and a clever matrix design, coupled with the zero potential method, proves that high-resolution plantar pressure sensing can be achieved at a very low cost.
Key Results
The instrumented shoe system demonstrated promising performance in gait phase detection:
- Detection Rate: 100% detection rate for all key gait events (HC, FC, HO, TO) and stance sub-phases (IC, FF, PO).
- Gait Event Detection Accuracy:
- Straight-line walking: Mean absolute differences (ADs) ranged from 1.9 to 2.8 data frames (38-56 ms).
- Turning walking: ADs ranged from 2.0 to 4.2 data frames (40-83 ms).
- Gait Phase Duration Estimation Accuracy:
- Straight-line walking: ADs ranged from 2.3 to 2.9 data frames (45-58 ms).
- Turning walking: ADs ranged from 2.6 to 3.9 data frames (51-77 ms).
- Overall Performance: The mean absolute time differences for gait phase detection were comparable to state-of-the-art systems, ranging from 45-58 ms for straight-line walking and 51-77 ms for turning walking.
Limitations and Future Work
The authors acknowledge several limitations and suggest future research directions:
- Swing Phase Duration: The study did not evaluate the swing phase duration due to limitations in the force plate system's ability to capture multiple consecutive stance phases. Future work could involve using modified force plate systems or alternative reference methods.
- Shoe Size and Participant Diversity: The study was conducted with only one shoe size (US 9) and healthy participants. Future research should investigate performance across different shoe sizes and include participants with pathological gaits to assess clinical relevance.
- Power Consumption: While the system is low-cost, increasing the number of sensing units for even higher resolution might increase power consumption, potentially limiting practical applications. Further research is needed to optimize the relationship between power consumption and sensor unit density.
- Manufacturing Consistency: The inherent variability in the FSF-tape-insole interface can introduce noise. Standardized manufacturing processes are recommended to improve detection accuracy.
- Data Output Rate: Increasing the data output rate beyond 50 Hz could further reduce detection errors. Processing data onboard and transmitting only key events wirelessly is a potential strategy.
Summary updated