❏ Interaction Estimation
The research outline is to estimate the movements of each person and the interaction movements of multiple people in a 2D moving video, such as walk, handshake, talk and etc..First, the 2D moving video is analyzed to estimate the 3D joint coordinates. Datasets are recorded and collected by using multiple cameras for the 2D video, and the dataset is converted to skeleton images in 3D space by a model called 3DPS. It also records the coordinates of each person’s skeleton. Based on this coordinated information, we can further determine the behavior of each person and the behavior of their interaction. For classification, I created my own dataset and created various labels containing each person’s own actions and interaction actions. I use these datasets to generate models that can be used to determine the behavior of each person and the behavior of possible interactions between people in real time. In the future, when autonomous driving becomes widespread, for example, if the bus doesn’t have a driver, my research can be used to quickly find accidents and problems. For the safety of public transport, we hope that this study can be used to ensure the safety of bus passengers.

Yiwen Fang