超広視野角一人称視点映像を用いた光学式動作計測システムに関する研究

点击翻译按钮获取中文摘要

Last year, we successfully implemented a system that estimates the multi-modal human motion using the ultra-wide fisheye camera. This year, we decided to expand the system's function to estimate the motion of third-person captured by the fisheye camera. To achieve this goal, a dataset was synthesized using Computer Graphics, and a new neural network pipeline was implemented using the synthesized data and evaluated through experiments. We submitted a paper describing the information obtained through this study and published it at the Augmented Humans 2022 international conference.Next, a study was conducted on easily creating mixed reality content based on the motion information obtained from these systems with simple user interaction. This study implemented a system that generates pseudo-2.5D content by fusion of motion and texture information estimated through neural networks and computer vision algorithms. This system's implementation method and experimental contents were described in a paper and published in an international academic journal.Finally, based on the studies conducted so far, I wrote the Ph.D. thesis on markerless motion capture and visualization methods and passed the degree examination.