Prediction of Three-Directional Ground Reaction Forces during Walking Using a Shoe Sole Sensor System and Machine Learning

We developed a shoe sole sensor system with four high-capacity, compact triaxial force sensors using a nitrogen added chromium strain-sensitive thin film mounted on the sole of a shoe. Walking experiments were performed, including straight walking and turning (side-step and cross-step turning), in six healthy young male participants and two healthy young female participants wearing the sole sensor system. A regression model to predict three-directional ground reaction forces (GRFs) from force sensor outputs was created using multiple linear regression and Gaussian process regression (GPR). The predicted GRF values were compared with the GRF values measured with a force plate. In the model trained on data from the straight walking and turning trials, the percent root-mean-square error (%RMSE) for predicting the GRFs in the anteroposterior and vertical directions was less than 15%, except for the GRF in the mediolateral direction. The model trained separately for straight walking, side-step turning, and cross-step turning showed a %RMSE of less than 15% in all directions in the GPR model, which is considered accurate for practical use.

INSOLE-BASED ESTIMATION OF COMPLETE GROUND REACTION FORCE WITH GAUSSIAN KERNEL REGRESSION AND DATA EXPANSION

A method for estimating ground reaction force (GRF) with plantar pressure was proposed in this paper. The estimation model was constructed to approximate the nonlinear relationships between GRF and the plantar pressure according to the linear combinations of Gaussian kernel functions. Partial least squares regression (PLSR) was adopted to obtain model parameters and eliminate multicollinearity among the pressure components. The general model and subject-specific models were constructed for 12 male and 4 female subjects. Moreover, a data expansion method was introduced for the establishment of subject-specific model, which is implemented by searching and adopting the data with consistent statistical characteristics in a pre-established database. That approach is particularly meaningful for the group whose walking ability is limited or clinic where the force platform is not available. The NRMSEs (%) for general model were 5.27–7.85% (GRF_V), 7.35–8.53% (GRF_ML), and 8.82–10.54% (GRF_AP). The maximum NRMSEs (%) for subject-specific models were 5.02% (GRF_V), 9.91% (GRF_ML), and 10.23% (GRF_AP). Results showed that both general and subject-specific models achieved higher accuracy than existing methods such as linear regression and neural network methods.

Estimation of Vertical Ground Reaction Force Using Low-Cost Insole With Force Plate-Free Learning From Single Leg Stance and Walking

For the evaluation of pathological gait, a machine learning-based estimation of the vertical ground reaction force (vGRF) using a low-cost insole is proposed as an alternative to costly force plates. However, learning a model for estimation still relies on the use of force plates, which is not accessible in small clinics and individuals. Therefore, this paper presents a force plate-free learning from a single leg stance (SLS) and natural walking measured only by the insoles. This method used a linear least squares regression that fits insole measurements during SLS to body weight in order to learn a model to estimate vGRF during walking. Constraints were added to the regression so that vGRF estimates during walking were of proper magnitude, and the constraint bounds were newly defined as a linear function of stance duration. Moreover, a lower bound for the estimated vGRF in mid-stance was added to the constraints to enhance estimation accuracy. The vGRF estimated by the proposed method was compared with force platforms for 4 healthy young adults and 13 elderly adults including patients with mild osteoarthritis, knee pain, and valgus hallux. Through the experiments, the proposed learning method had a normalized root mean squared error under 10% for healthy young and elderly adults with stance durations within a certain range (600-800 ms). From these results, the validity of the proposed learning method was verified for various users requiring assessment in the field of medicine and healthcare.

Development of a Bendable Outsole Biaxial Ground Reaction Force Measurement System

Wearable ground reaction force (GRF) measurement systems make it possible to measure the GRF in any environment, unlike a commercial force plate. When performing kinetic analysis with the GRF, measurement of multiaxial GRF is important for evaluating forward and lateral motion during natural gait. In this paper, we propose a bendable GRF measurement system that can measure biaxial (vertical and anterior-posterior) GRF without interrupting the natural gait. Eight custom small biaxial force sensors based on an optical sensing mechanism were installed in the proposed system. The interference between two axes on the custom sensor was minimized by the independent application of a cantilever structure for the two axes, and the hysteresis and repeatability of the custom sensor were investigated. After developing the system by the installation of force sensors, we found that the degree of flexibility of the developed system was comparable to that of regular shoes by investigating the forefoot bending stiffness. Finally, we compared vertical GRF (vGRF) and anterior-posterior GRF (apGRF) measured from the developed system and force plate at the same time when the six subjects walked, ran, and jumped on the force plate to evaluate the performance of the GRF measurement system.

Insole-Based Estimation of Vertical Ground Reaction Force Using One-Step Learning With Probabilistic Regression and Data Augmentation

An insole-based estimation of the vertical ground reaction force (vGRF) is proposed as an alternative to costly force plates for the evaluation of pathological gait. However, machine learning techniques for estimation still rely on the use of force plates. Moreover, measuring plural walking steps in order to prevent overfitting induces fall risks and physically taxes the patients. Therefore, this paper presents an accessible and efficient learning scheme for the insole-based estimation of vGRF. In this system, we employ a low-cost scale as an alternative to force plates. Then, we use Gaussian process regression (GPR) to learn a model in order to estimate vGRF without overfitting of small-sized data sets corrupted by measurement errors and noise of the devices. In addition, we propose a "one-step learning" scheme based on a probabilistic data augmentation. This approach augments actual measurements of a minimum (just one) walking step to a virtual data set for plural steps by considering their typical variability between steps. In experiments, the GPR models learned from two walking steps estimated vGRF with mean errors of 8% or under for entire/local magnitudes. Moreover, the learning from one step with probabilistic augmentation enhanced the estimation accuracy.

Development of A Textile Capacitive Proximity Sensor and Gait Monitoring System for Smart Healthcare

Gait is not only one of the most important functions and activities in daily life but is also a parameter to monitor one's health status. We propose a single channel capacitive proximity pressure sensor (TCPS) and gait monitoring system for smart healthcare. Insole-type TCPS (270 mm in length) was designed consisting of three layers including two shield layers and a sensor layer. Analyzing the step count and stride time are the basic indicators in gait analysis, thus they were selected as evaluation indicators. A total of 12 subjects participated in the experiment to evaluate the resolution of our TCPS. To evaluate the accuracy of TCPS, step count and its error rates were simultaneously detected by naked eye, ZIKTO Walk (ZIKTO Co., Korea), and HJ-203-K pedometer (Omron Co., Japan) as reference. Results showed that the error rate of 1.77% in TCPS was lower than those of other devices and correlation coefficient was 0.958 (p-value = 0.000). ZIKTO Walk and pedometer do not provide information on stride time, therefore it was detected by F-scan (Tekscan, USA) to evaluate the performance of TCPS. As a result, error rate of stride time measured by TCPS was found to be 1% and the correlation coefficient was 0.685 (p-value = 0.000). According to these results, our proposed system may be helpful in development of gait monitoring and measurement system as smart healthcare.