Comparison of the Accuracy of Ground Reaction Force Component Estimation between Supervised Machine Learning and Deep Learning Methods Using Pressure Insoles

Abstract

The three Ground Reaction Force (GRF) components can be estimated using pressure insole sensors. In this paper, we compare the accuracy of estimating GRF components for both feet using six methods: three Deep Learning (DL) methods (Artificial Neural Network, Long Short-Term Memory, and Convolutional Neural Network) and three Supervised Machine Learning (SML) methods (Least Squares, Support Vector Regression, and Random Forest (RF)). Data were collected from nine subjects across six activities: normal and slow walking, static with and without carrying a load, and two Manual Material Handling activities. This study has two main contributions: first, the estimation of GRF components (Fx, Fy, and Fz) during the six activities, two of which have never been studied; second, the comparison of the accuracy of GRF component estimation between the six methods for each activity. RF provided the most accurate estimation for static situations, with mean RMSE values of RMSE_Fx = 1.65 N, RMSE_Fy = 1.35 N, and RMSE_Fz = 7.97 N for the mean absolute values measured by the force plate (reference) RMSE_Fx = 14.10 N, RMSE_Fy = 3.83 N, and RMSE_Fz = 397.45 N. In our study, we found that RF, an SML method, surpassed the experimented DL methods.

Keywords: GRF component estimation; deep learning; force plate measurement; insole pressure measurement; manual material handling; supervised machine learning; walking activities.

Figure 3

The subject goes down and up the plate for each trial following the black arrow trajectory for normal and slow walking.

Figure 4

Static situation (left image), static situation with a 5 kg load (right image).

Figure 5

The participant starts by carrying a 5 kg load from the bottom (chair) (left image) to the top (table) (right image) and vice versa.

Figure 6

The participant starts by carrying a 5 kg load from the far left of the table (left image) to the far right of the table (right image) and vice versa.

Figure A1

The curves of the three GRF components for a “slow walk” step (top image), a trial of “bottom-top with CL” (middle image) and “left-right with CL” (bottom image) of the test dataset for the right foot using the RF method for intras strategy. The measured curves are presented in blue and the estimated curves are presented in red.

Figure 1

Flow chart for GRF component estimation from PP data. PP training dataset is utilized for DL/SML modeling, incorporating corresponding GRF force plate data (depicted in red). PP testing dataset is then employed to evaluate the performance of GRF components using the corresponding GRF force plate data (shown in green).

Figure 2

The location of the 16 pressure sensors along the insole.

Figure 7

An example of synchronization for an excerpt of walking activity with the Moticon insole for the right foot.

Figure 8

The curves of mean RMSE between the estimated (by insole PP data) and measured (by force plate) GRF components of the right foot for each DL and SML method for the test dataset, covering both strategies and each activity. The black curve represents the “normal walk”, the magenta curve represents the “slow walk”, the red curve represents the “static situation”, the blue curve represents the “static situation with CL”, the green curve represents “bottom-top with CL”, and the cyan curve represents “left-right with CL”. The mean R values are indicated on the RMSE values for each DL and SML method.

Figure 9

The curves of mean RMSE between the estimated (by insole PP data) and measured (by force plate) GRF components of the left foot for each DL and SML method for the test dataset, covering both strategies and each activity. The black curve represents the “normal walk”, the magenta curve represents the “slow walk”, the red curve represents the “static situation”, the blue curve represents the “static situation with CL”, the green curve represents “bottom-top with CL”, and the cyan curve represents “Left-right with CL”. The mean R values are indicated on the RMSE values for each DL and SML method.

Figure 10

The curves of Fz estimated for both feet (the solid blue and green lines for the right and left foot, respectively) and measured (the solid red and black lines for the right and left foot, respectively) using the RF method for samples from the test dataset for the intras strategy. The curves of the summation of Fz for both feet estimated are presented by the dashed green line and those measured are presented by the dashed red line. The curves of the subject’s weight for the static situation or the subject’s weight plus an additional 5 kg for the static situation with CL are presented in the dashed black line. The right image pertains to the static situation with CL and the left image pertains to the static situation. In both images, g is the gravity, which is equal to 10 m/s².