Week 2: Data Pipeline

Day 10: Pose and Keypoint Extraction

Day 10 of 2818 minGoal - Learn - Example - Practice - Checkpoint

Goal

Understand pose/keypoint extraction in plain English.

Learn

  • Pose extraction turns each frame into coordinates for named body parts. These named dots are keypoints.
  • For sign language, useful keypoint groups often include shoulders, elbows, wrists, hands, fingers, face, eyes, mouth, nose, and sometimes torso or feet.
  • Tools such as DWPose, OpenPose, MediaPipe, and other estimators can produce keypoints, but each has its own strengths and failure modes.

Example

  • A frame may store left shoulder, left elbow, left wrist, and 21 left-hand points. A hand point might include x, y, and confidence.
  • If confidence drops during fast motion, the pose preview may show fingers jumping or disappearing.

Practice

  1. On a sample image, mark shoulders, elbows, wrists, fingertips, eyes, nose, and mouth.
  2. Write which points are most important for ASL clarity in that frame.

Checkpoint

Before moving on

You can explain keypoints as named dots on the body that change over time.

Quality note

Quality note

For ASL, hand and face keypoints usually need closer inspection than large body joints because small errors can change meaning.