Activity Recognition with Loose Clothing

Abstract

With the advancement in electronic textile and on-body wearable sensing technologies, human activity recognition has gained substantial research and is becoming an area of significance. The integration of sensors into garments has paved the way for activity recognition, enabling users to engage in extended human motion recordings. Recent novel insights shed light on the revelation that sensors attached on clothing exhibit higher activity recognition than sensors attached on the rigid part of an object in motion. In which fuelled interest towards the exploration the different parameters of fabric on the activity recognition. Fabric is tested while in periodic motion using a manually built scotch yoke, and the KUKA robot manipulator. The paper in hand reports the improved frequency classification of clothing-attached sensors with fabric in perpendicular orientation, triple layer, and large width. The amplified statistical distance between rigid and clothing- attached sensors due to more fabric layers is also conveyed. Furthermore, plane classification with fabric was explored, and clothing-attached sensors showed higher classification accuracy than rigid-attached sensor. These findings help refine the probabilistic model of fabric motion that was introduced in recent studies and contribute to the growing field of human activity recognition.