Skip to main content
Skip to main content
Intermediate

Pose Estimation

Also known as: human pose detection, body keypoint detection

Definition

Pose estimation is the computer-vision process that detects the positions of major body joints (keypoints) in each video frame. This produces the skeleton that SwingVantage uses to measure angles and movement patterns.

Pose estimation runs a neural network over video frames to locate landmarks like shoulders, elbows, wrists, hips, knees, and ankles. It runs on a sample of frames from your clip and is the foundation for all motion metrics. Accuracy depends on visibility: occluded joints (hidden by clothing, body rotation, or poor lighting) produce lower-confidence keypoints that are labeled accordingly. Motion Lab runs pose estimation on your device, so the video file does not leave it.

Tips by skill level

Beginner tip

You don't need to understand how pose estimation works — just know that good video setup (light, angle, frame rate) directly improves its accuracy. This improves the reliability of everything else.

Advanced note

Review the skeleton overlay before trusting any metric. If a joint is visibly drifting off your body at a key moment (like impact), the metric derived from that joint at that frame should be treated as an estimate, regardless of the stated confidence.

Example

Pose estimation places 33 body landmarks on each sampled frame of a golf swing, letting the engine compute hip-to-shoulder separation angle through impact.

Why it matters

Without reliable pose estimation, angles and timing metrics cannot be computed at all. It is the step where good video setup pays off most.

How it shows up on video

The skeleton overlay is the visible output of pose estimation — when estimation is working well, joints track smoothly. When it fails, landmarks jump or disappear, usually around fast-moving limbs or occluded joints.

Common mistakes

  • Treating all pose estimation output as equally reliable — landmark confidence varies by joint, frame, and video quality.
  • Assuming the system always knows where your wrist is during a fast swing — fast-moving distal joints are the hardest to track reliably.
  • Wearing clothing that matches the background color, which causes landmarks to drift.
  • Not checking the skeleton overlay at all — it is the most direct way to verify that tracking was accurate before trusting any analysis finding.

In SwingVantage Motion Lab

Motion Lab runs pose estimation on a sample of frames from your clip before building its 3D view. Short gaps between tracked frames are filled in, and the 3D view labels a tracking gap or an illustrated pose rather than hiding it.

Still have a question about Pose Estimation? Ask SwingVantage

See it in a real analysis

Try SwingVantage on one of your own swings — free to start.