A Method for Detecting Human-Object Interaction Using RGB-D Video
Taichi Uchimura, Toru Abe, Takuo Suganuma · 2024
The action of a person moving an object with their hands plays a central role in human-object interaction. In this article, we propose a novel method to detect the action of moving an object with the hands in RGB-D video. Our method uses 2D skeleton information and 3D motion (scene) flow obtained from RGB-D video to analyze the movements around the hand, and determines that the hand moves an object when movements similar to those of the forearm occur around the hand. By effectively utilizing different information available from RGB-D video, this method is expected to be able to efficiently detect actions of moving objects with hands in various directions.