[2E2] AR / MR Evaluation and Embodiment

[2E2-05] Body motion forecast from 3D motion sequences by using neural network.

○Yuuki Horiuchi¹, Yasutishi Makino¹, Hiroyuki Shinoda¹ (1. The University of Tokyo)

In this paper, we propose a machine learning-based system that can forecast body motion 0.5 seconds before. The system detects 25 human body joints by Kinect V2. We created 6-layered neural network system for machine-learning. We achieved realtime motion forecast system. This can be used to estimate human gesture inputs in advance, instruct sports actions properly, and prevent elderly from falling to the ground, and so on.