In order to operate a gait rehabilitation device, it is necessary to accurately classify the states appearing in activities of daily living (ADLs). In the case of force sensing resistors (FSRs), which are often used as pressure sensors in gait analysis, it is desirable to replace them with other sensors because of their low durability. In the present study, capacitive-type pressure sensors, as an alternative to FSRs, were developed, and their performance was evaluated. In addition, the timed up and go test was performed to measure the ground reaction force in healthy individuals, and a machine learning technique was applied to the calculated biosignal parameters for the classification of five types of ADLs. The performance evaluation results showed that a sensor with thermoplastic polyurethane (substrate and dielectric layer material) and multiwall carbon nanotubes (conductive layer) has sufficient sensitivity and durability for use as a gait analysis pressure sensor. Moreover, when an overlapping filter was applied to the four-layer long short-term memory (LSTM) or the five-layer LSTM model developed for motion classification, the precision was greater or equal to 95%, and unstable errors did not occur. Therefore, when the pressure sensor and ADLs classification algorithm developed in this study are applied, it is expected that motion classification can be completed within a time range that does not affect the control of the gait rehabilitation device.

Gait assessment is an important tool for determining whether a person has a gait disorder. Existing gait analysis studies have a high error rate due to the heel-contact-event-based technique. Our goals were to overcome the shortcomings of existing gait analysis techniques and to develop more objective indices for assessing gait disorders. This paper proposes a method for assessing gait disorders via the observation of changes in the center of pressure (COP) in the medial–lateral direction, i.e., COPx, during the gait cycle. The data for the COPx were used to design a gait cycle estimation method applicable to patients with gait disorders. A polar gaitogram was drawn using the gait cycle and COPx data. The difference between the areas inside the two closed curves in the polar gaitogram, area ratio index (ARI), and the slope of the tangential line common to the two closed curves were proposed as gait analysis indices. An experimental study was conducted to verify that these two indices can be used to differentiate between stroke patients and healthy adults. The findings indicated the potential of using the proposed polar gaitogram and indices to develop and apply wearable devices to assess gait disorders.

Stride length (SL), foot clearance (FC), and foot progression angle (FPA) are the key parameters for diagnosing gait disorders. This study used the distance data between two feet measured by ultra-wideband (UWB) sensors installed on shoes and proposed a method for estimating the three gait parameters. Here, a method of compensating the offset of the UWB sensor and estimating the distances between a base sensor installed on one foot during the stance phase and three UWB sensors on the other during the swing phase was applied. Foot trajectory was acquired in a gait experiment with ten healthy adults walking on a treadmill. The results were compared with those obtained using a motion capture system (MCS). The UWBs sensor displayed average errors of 45.84 mm, 7.60 mm, and 2.82° for SL, FC, and FPA, respectively, compared with the MCS. A similar accuracy level was achieved in a previous study that used an inertial measurement unit (IMU). Thus, these results suggest that UWB sensors can be extensively applied to sensor systems used to analyze mobile gait systems.