FlexiMoCap is a multi-RGBD camera system that tracks a person’s motion in 3D in real time. The student will build a pipeline that connects 3–6 depth cameras and outputs one unified 3D skeleton (joint positions over time) in a single world coordinate system. Compared to using one camera, multiple cameras can see the body from different angles, which reduces missing joints caused by occlusion and makes tracking more stable. The system should also support “add a camera anytime”: if a new camera is plugged in, it can be included in the system without rebuilding everything from scratch.
The project has two main tasks. (1) Fast calibration: implement a quick procedure (using a printed marker board / calibration board) to estimate each camera’s pose (position + orientation) and finish calibration in minutes. The system should also show a simple calibration quality indicator (e.g., error value and “ready / not ready”). (2) Camera placement study: test several real camera layouts (different number of cameras, heights, angles, and distances) and evaluate which setups work best. The student will report results using clear metrics such as joint detection rate, 3D jitter (stability), and tracking drop rate, and summarize practical “best-practice” placement guidelines. A pose model (OpenPose-style ,yolo,keypoints) may be used as a tool for detecting failures and matching the same person across different camera views, improving the final 3D tracking output.