Skip to main content
Skip to main content
Intermediate

Landmark Tracking

Also known as: keypoint tracking, joint tracking

Definition

Landmark tracking follows the position of each detected body keypoint across consecutive video frames. This creates a time-series trajectory for every joint that enables timing and velocity measurements.

Pose estimation finds where joints are in a single frame; landmark tracking connects those positions over time to produce trajectories. From trajectories you can derive velocities, accelerations, angles at specific moments (like impact), and sequencing relationships between body segments. Tracking errors — where a joint jumps between frames — are a known challenge and are handled by filtering algorithms. Any metric derived from a noisy trajectory is labeled with lower confidence or flagged as an estimate.

Tips by skill level

Beginner tip

You don't need to analyze trajectories directly. Just check the final skeleton overlay for obvious drifting joints — the system automatically handles the trajectory math behind the scenes.

Advanced note

When a joint's path looks odd, step through those frames in slow motion. A smooth path that stays on your body is likely real movement; a one-frame jump off the body is likely a tracking error.

Example

Landmark tracking follows your hands from the top of a golf swing to impact. That path is what the hand trace in your Motion Lab results draws.

How it shows up on video

Play the video with the skeleton overlay at reduced speed — landmark tracking is visible as the smooth (or not-so-smooth) path each joint follows across frames. Abrupt jumps indicate tracking errors that will corrupt derived metrics.

Common mistakes

  • Assuming all joints track equally well. The wrist and elbow are significantly harder to track than the hip or shoulder due to their speed and size.
  • Using a metric derived from a noisy trajectory without checking the trajectory plot first.
  • Confusing a tracking jump with a real movement — if a landmark jumps a long way in a single frame, it is almost certainly an error.
  • Not understanding that velocity and timing metrics are more sensitive to tracking errors than position metrics. A small position error becomes a large velocity error when differentiated over time.

In SwingVantage Motion Lab

Motion Lab draws the path of your hands, or of the implement when it can be tracked, over your clip. The 3D view labels the moments where tracking dropped out.

Still have a question about Landmark Tracking? Ask SwingVantage

See it in a real analysis

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