A Motion-Simulation Platform to Generate Synthetic Motion Data for Computer Vision Tasks

A. Chalmers , J. Zhao , W. K. Hoh , J. J. Drown , S. Finnie , R. Yao , J. Lin , J. Wilmott , A. Dey , M. Billinghurst , T. Rhee

SIGGRAPH Asia Technical Communications

Motion-related computer-vision tasks need large amounts of intrinsic motion data that is costly to capture and label in the real world. This paper presents a Motion-Simulation Platform, built inside a game engine, that extracts both RGB imagery and the corresponding motion field for any scene. It is extendable to new scenes and supports both user-driven and AI-bot navigation, enabling large-scale automated collection of synthetic motion data for training vision models.

@inproceedings{chalmers2023,
  author    = {A. Chalmers and J. Zhao and W. K. Hoh and J. J. Drown and S. Finnie and R. Yao and J. Lin and J. Wilmott and A. Dey and M. Billinghurst and T. Rhee},
  title     = {A Motion-Simulation Platform to Generate Synthetic Motion Data for Computer Vision Tasks},
  booktitle = {SIGGRAPH Asia Technical Communications},
  pages     = {21:1-21:4},
  doi       = {10.1145/3610543.3628795},
  year      = {2023}
}