VSLink: A Fast and Pervasive Approach to Physical Cyber Space Interaction via Visual SLAM
Han Zhou,Jiaming Huang,Hongchang Fan,Geng Ren,Yi Gao,Wei Dong
DOI: https://doi.org/10.1109/MSN57253.2022.00054
2022-01-01
Abstract:With the fast growth of the Internet of Things, people now are surrounded by plenty of devices. To achieve efficient interaction with these devices, human-device interaction technologies are evolving. Because existing methods (mobile App) require users to remember the mapping between the real-world device and the digital one, an important point is to break such a gap. In this paper, we propose VSLink, which offers human-device interaction in an Augmented-Reality-like manner. VSLink achieves fast object identification and pervasive interaction for fusing the physical and cyberspace. To improve processing speed and accuracy, VSLink adopts a two-step object identification method to locate the interaction targets. In VSLink, visual SLAM and object detection neural networks detect stable/-movable objects separately, and detection prior from SLAM is sent to neural networks which enables sparse-convolution-based inference acceleration. VSLink offers a platform where the user could customize the interaction target, function, and interface. We evaluated VSLink in an environment containing multiple objects to interact with. The results showed that it achieves a 33% network inference acceleration on state-of-the-art networks, and enables object identification with 30FPS video input.