Chia-Hui Lin’s research while affiliated with National Taipei University of Technology and other places

Previous works have shown the potential of immersive technologies to make physical activities a more engaging experience. With encountered-type haptic feedback, users can perceive a more realistic sensation for exertion interaction in substitutions reality. Although substitutional reality has utilized physical environments, props, and devices to provide encountered-type haptic feedback, these cannot withstand the fierce force of humans and do not give feedback when users move around simultaneously, such as in combat sports. In this work, we present Hit Around, a substitutional moving robot for immersive and exertion interaction, in which the user can move and punch the virtual opponent and perceive encountered-type haptic feedback anywhere. We gathered insight into immersive exertion interaction from three exhibitions with iterative prototypes, then designed and implemented the hardware system and application. To understand the ability of mobility and weight loading, we conducted two technical evaluations and a laboratory experiment to validate the feasibility. Finally, a field deployment study explored the limitations and challenges of developing immersive exertion interaction with encountered-type haptics.

We present a self-trainable method, Mask2Hand, which learns to solve the challenging task of predicting 3D hand pose and shape from a 2D binary mask of hand silhouette/shadow without additional manually-annotated data. Given the intrinsic camera parameters and the parametric hand model in the camera space, we adopt the differentiable rendering technique to project 3D estimations onto the 2D binary silhouette space. By applying a tailored combination of losses between the rendered silhouette and the input binary mask, we are able to integrate the self-guidance mechanism into our end-to-end optimization process for constraining global mesh registration and hand pose estimation. The experiments show that our method, which takes a single binary mask as the input, can achieve comparable prediction accuracy on both unaligned and aligned settings as state-of-the-art methods that require RGB or depth inputs.