Visual Cone Detection (and pose estimation)

Tentative Plans

• Train YOLO v11 on cone dataset (Roboflow does this for us)

• Steal cone images from cone dataset by cropping out the given bounding boxes

• Label many keypoints on Roboflow

• Train CNN to obtain keypoints

• (RANSAC) PnP on keypoints to infer cone pose

• Testing/ROS integration

• Profit!

File Structure

• monocular: keypoint calculation and PnP

• stereo: stereo

• util: miscellaneous helper scripts