Motion Classification With Embroidery Bend Sensors Using Multiple Zigzag-Stitch for Loose-Fitting Garments

Abstract

This study proposes a wearable motion classification system by employing commercially available conductive threads and everyday garments. The unique feature of the proposed system is an embroidery bending sensor that does not necessitate tight-fitting with the body, which contrasts with traditional motion sensing systems. Therefore, integration can be simplified by allowing motion classification onto loose-fitting everyday garments. The sensor that exhibits a change in resistance to bending deformation is realized by applying multiple zigzag stitch to the fabric. In addition, a significant resistance change is realized during the phase transition of motions (e.g., stance and swing phase). Therefore, motion can be categorized without making joint angle measurements. We fabricated a prototype by attaching the sensors to a commercially available work jacket and pants by sewing them onto the fabric. Ten participants were requested to perform ten various activities (e.g., walking, jogging, ascending and descending stairs). The findings demonstrated that the sensor can measure the degree of joint flexion, the flexion cycle, and the timing of flexion during the wearer's activities. Moreover, motion classification was performed by training a one-dimensional convolutional neural network (1D-CNN) model with the sensor signals. The model successfully learned the differences in signal amplitude and frequency as distinguishing features of each activity, resulting in an average classification accuracy of 99.02% across the ten types of activities.