Gretchen — 2DOF Gimbal Object Detection Robot
3D-printed 2DOF gimbal robot built at Seoul National University — uses color filtering, coordinate transforms, and YOLOv3 to detect, label, and track objects in real time.
Gretchen is a 3D-printed 2DOF gimbal robot built as part of the Seoul National University robotics program. The project combined classical computer vision with YOLOv3 for real-time object detection and tracking.
The vision pipeline uses color filtering to isolate regions of interest and edge detection to define object boundaries. A calibrated transform between 2D image coordinates and 3D world space allows the gimbal to track objects through space as they move.
For detection and labeling, YOLOv3 was applied to identify objects in real time. Custom model weights were trained in Jupyter Lab using manually acquired and annotated images — targeting five classes of surgical instruments, achieving approximately 60% confidence from a dataset of 300 images. Using YOLOv3's built-in class set, the system successfully tracked a ball through space and live-detected over 50 distinct object types.
