A Monocular Ranging Method in Floating Touch System

The field of human-computer interaction based on floating touch is an emerging research field. While exist-ing floating touch systems rely on high-cost hardware (e.g., LiDAR, depth cameras), low-cost monocular camera-based solutions remain understudied. This paper presents a monocu-lar ranging method in floating touch system. Firstly, a shallow neural network is used to fit an empirical function to establish a hand distance model, and a deep neural network proposes a distance estimation algorithm based on camera pose estima-tion. Then, a data fusion strategy is used to integrate the two to construct a multi-layer network that effectively improves the accuracy of monocular ranging for hands. Finally, a vir-tual hand-based pose compensation algorithm is introduced, significantly enhancing the overall method’s precision and robustness in complex hand movements or angles. Experi-ments show that the ranging algorithm based on camera pose estimation can calculate the camera’s offset components in all directions, addressing issues like pose angles confusion in monocular distance measurement. The pose compensa-tion algorithm based on a virtual hand model can extract virtual coordinates and specific part parameter data, calcu-lating compensation errors and improving the accuracy and robustness of hand distance measurement algorithms. The proposed monocular ranging method outperforms monocular metric depth estimation methods in outdoor and occlusion conditions, achieving stable millimeter level accuracy in the range of 20-100 cm. Compared with traditional distance measurement methods, the mean relative error is reduced by 15.45%.