Hand pose estimation plays an important role in applications such as human-computer interaction (HCI), smart healthcare, robotics. We present SleevePose, a wearable system composed of an elastic, resistive pressure sensing sleeve that infers hand motion by capturing pressure changes on the forearm surface. SleevePose consists of two fabric layers with conductive stripes formed using a double-jersey jacquard knitting technique: horizontal on the outer layer and vertical on the inner layer. Their orthogonal intersection forms a 13×9 resistive pressure sensing matrix. To validate the system's effectiveness, we design a neural network combing a ResNet backbone and Transformer encoder, which takes temporal pressure sequences as input and predicts the parameters of MANO (hand Model with Articulated and Non-rigid defOrmations). We conduct a user study involving 18 users, collecting over 720k frames of synchronized pressure and RGB data across various scenarios. In the user-dependent setting, our system achieves a mean per joint position error (MPJPE) of 14.04 mm. Our results highlight the potential of flexible textile sensors for accurate hand pose estimation.
Niu et al. (Mon,) studied this question.