Markerless Motion Capture
Also known as: Marker-free tracking, Vision-based motion capture, Deep learning pose estimation
Markerless motion capture infers the 3D positions and joint angles of a moving subject from video sequences using computer vision and machine learning. Pioneered by deep learning approaches such as OpenPose and MediaPipe, it eliminates the need for reflective markers or inertial sensors, making motion capture accessible and practical for real-world applications.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use markerless motion capture when markers are impractical (large crowds, outdoor settings, water immersion) or when rapid deployment is required. It excels in fitness, sports analysis, clinical screening, and entertainment. Assumptions include sufficient image resolution to resolve joints, adequate lighting, and pre-training on diverse poses. Be cautious with heavy occlusion, fast motion blur, or extreme poses outside the training distribution.
Strengths & limitations
- No specialized hardware or markers required; uses standard cameras
- Practical for real-time applications and field deployments
- Handles multiple people in a single view
- Improving accuracy as deep learning models mature and datasets expand
- Accuracy degrades with occlusion, motion blur, and poor lighting
- Lower spatial resolution than marker-based systems in clinical settings
- Temporal noise and jitter without post-processing or smoothing
- Struggles with extreme poses or subjects outside the training distribution
Frequently asked
How accurate is markerless motion capture compared to marker-based systems?
In controlled laboratory settings with good lighting, state-of-the-art markerless systems achieve 2–5 cm 3D error at major joints; marker-based systems achieve sub-centimeter accuracy. Markerless degrades more in unconstrained environments.
Can I improve accuracy with multi-view markerless capture?
Yes. Multiple synchronized cameras allow 3D triangulation and reduce ambiguity from self-occlusion. Accuracy improves especially when combining 2D detections from different viewpoints.
How do I reduce temporal jitter in markerless tracking?
Apply temporal smoothing (Kalman filtering, butterworth filtering, or optical flow) post-hoc to the joint trajectories. Some frameworks include smoothing as a built-in option.
Sources
- Cao, Z., Simon, T., Wei, S. E., & Sheikh, Y. (2017). Realtime multi-person 2D pose estimation using part affinity fields. IEEE Conference on Computer Vision and Pattern Recognition (CVPR). DOI: 10.1109/CVPR.2017.143 ↗
- Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep Learning. MIT Press. link ↗
How to cite this page
ScholarGate. (2026, June 3). Markerless Motion Capture. ScholarGate. https://scholargate.app/en/biomechanics/markerless-motion-capture
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- DTW Gait AnalysisBiomechanics↔ compare
- Forward KinematicsBiomechanics↔ compare
- Inverse DynamicsBiomechanics↔ compare