Written and maintained by CASRAI Editorial Board
Published
On this page
Short answer
OpenCap is an open-source, cloud-based platform developed at Stanford University that computes three-dimensional human kinematics and musculoskeletal dynamics from synchronized smartphone videos. Pairing two iOS devices with deep learning pose estimation and physics-based OpenSim modeling, OpenCap estimates joint angles, moments, and muscle forces without optical markers or laboratory force plates.
Definition and scope
OpenCap is an open-access biomechanics platform developed by the Stanford Neuromuscular Biomechanics Laboratory (NMBL), the Mobilize Center, and the Wu Tsai Human Performance Alliance. It democratizes quantitative movement analysis by eliminating the cost and space constraints of traditional optical motion capture laboratories.
Conventional gait labs rely on infrared cameras, reflective skin markers, and floor force plates costing over $150,000. OpenCap replaces this infrastructure with two or more iOS devices on tripods. Videos stream to cloud servers that extract metric 3D kinematics and internal musculoskeletal forces.
The system supports clinical movement screening, athletic tracking, and research across disciplines represented by the Biomedical Engineering Society (BMES). In alignment with open-science standards highlighted by the Journal of Open Research Software, OpenCap provides transparent processing pipelines and shared anatomical models for reproducible movement science.
How it works or how it is used in practice
OpenCap operates through an automated four-stage pipeline integrating multi-view computer vision, neural networks, and OpenSim musculoskeletal simulations.
1. Spatial Calibration
Two iOS devices on tripods are positioned 1.5 to 3 meters apart, angled 30° to 45° off the progression axis to prevent self-occlusion. A 15-parameter OpenCV pinhole camera model loads pre-calibrated intrinsic optical parameters from a hardware database. Extrinsic parameters are solved by placing a 210×175 mm checkerboard on rigid plexiglass in the shared camera view. Tripods remain fixed thereafter.
2. 2D Pose Estimation
During movement trials, synchronized devices record video at 60 Hz. The web interface streams footage to cloud servers where High-Resolution Net (HRNet) detects 20 body keypoints per frame, including distal foot markers. Each keypoint receives 2D coordinates and a confidence score between 0 and 1.
3. Confidence-Weighted Triangulation and LSTM Marker Expansion
OpenCap reconstructs 3D keypoints via Direct Linear Transformation (DLT), weighting each camera view by keypoint confidence. Because 20 video keypoints cannot fully resolve joint rotations, two bidirectional Long Short-Term Memory (LSTM) networks predict 3D coordinates for 43 anatomical markers from the 20 triangulated points, reconstructing standard virtual marker trajectories.
4. OpenSim Musculoskeletal Scaling and Dynamic Simulation
A brief standing trial captures neutral posture. The OpenSim Scale tool scales a generic musculoskeletal model (Lai et al., 2017; Uhlrich et al.) featuring 33 degrees of freedom and 80 muscle-tendon compartments to subject anthropometry. OpenSim Inverse Kinematics computes joint angles. Muscle-driven tracking simulations then solve optimal control formulations to estimate ground reaction forces, joint moments (such as knee adduction and extension moments), and muscle activations without physical force plates.
Types or related concepts
Human motion capture spans four primary modalities, balancing precision, setup time, and equipment expense.
| Modality | Hardware Required | Kinematic Accuracy (RMSE) | Kinetic Capacity | Setup & Capture Time |
|---|---|---|---|---|
| Optical Marker Mocap | 8–16 IR cameras, markers, force plates, workstation | Gold standard (<1.5° instrument error; up to 8° skin artifact) | Direct ground reaction forces via floor force plates; inverse dynamics | 45–90 min per subject |
| Wearable IMU Arrays | 7–17 body sensors, straps, wireless receiver hub | 2.0°–12.0° RMSE (subject to magnetic drift) | Indirect estimation via machine learning or physics approximations | 15–30 min per subject |
| Multi-Camera Markerless (OpenCap) | 2+ iOS devices, tripods, checkerboard grid, web app | 2.0°–10.2° RMSE (mean rotational MAE 4.5°) | Physics-based muscle-driven tracking simulation without force plates | <5 min setup; automated cloud processing |
| Monocular Video Pose | Single RGB smartphone camera, 2D pose detector | 5.0°–15.0° RMSE (depth ambiguity along line of sight) | Primarily 2D planar kinematics; unconstrained 3D kinetics limited | Immediate recording; variable processing |
Wearable IMUs permit unconfined capture but lack global position reference. Optical mocap remains the reference standard for surgical planning but imposes strict throughput limits. Multi-camera OpenCap offers validated 3D metric kinematics and dynamics across portable environments.
Where it matters for labs, clinics or supply decisions
OpenCap shifts workflow organization, facility footprint, and software management across clinical and research environments.
Clinical Gait and Screening Throughput
In a 100-subject field study by Stanford researchers, a clinician using OpenCap collected and processed movement dynamics 25 times faster than laboratory optical motion capture, at less than 1% of the equipment cost. This throughput enables objective screening for ACL injury risks, post-operative squat symmetry evaluations, and osteoarthritis progression monitoring.
Data Governance and Multimodal Biosensors
OpenCap exports native OpenSim .trc and .mot files. Adhering to standards supported by Better Scientific Software (BSSw), these plain-text formats support long-term archival without vendor lock-in. Researchers can synchronize kinematic files with physiological streams, including electromyography or optical photoplethysmography sensors, supporting multidisciplinary biomechanics published in Medical & Biological Engineering & Computing.
Facility and Procurement Considerations
Adopting OpenCap alters laboratory capital expenditure:
- Space Requirements: No dedicated dark laboratory is required; any clear 4×4 meter space with uniform lighting suffices.
- Hardware Footprint: Capital expense shifts from proprietary infrared cameras to consumer smartphones and tripods.
- Consumables Elimination: Recurring purchasing of disposable reflective markers, adhesive collars, and skin-prep supplies is eliminated.
Common misunderstandings
- Misunderstanding: OpenCap is merely a 2D pose overlay.
Reality: OpenCap integrates confidence-weighted DLT triangulation, LSTM landmark expansion, and OpenSim dynamic simulations to compute 3D moments, contact loads, and muscle activations. - Misunderstanding: Markerless tracking cannot match optical marker accuracy.
Reality: OpenCap achieves a mean rotational MAE of 4.5° (RMSE 2.0°–10.2°). Because skin-mounted markers experience soft-tissue motion artifacts introducing up to 8° of error relative to bone pins, OpenCap performs within standard optical mocap error bounds. - Misunderstanding: Ground reaction forces require floor force plates.
Reality: OpenCap uses muscle-driven tracking simulations in OpenSim, solving optimal control problems that concurrently estimate ground reaction forces and internal joint loads from kinematics. - Misunderstanding: A single smartphone provides complete 3D dynamics.
Reality: Multi-camera OpenCap requires at least two cameras offset 30° to 45°. Single-camera setups suffer from line-of-sight depth ambiguity during non-sagittal movement.
FAQ
What equipment is required to run OpenCap?
OpenCap requires two or more iOS devices (iPhones or iPads), two tripods with smartphone mounts, a printed 210×175 mm checkerboard grid mounted on plexiglass, and a computer or tablet running a web browser.
How does OpenCap perform camera calibration without a wand?
OpenCap loads pre-computed intrinsic parameters for iOS cameras from an internal database. Placing the checkerboard pattern in the shared field of view allows OpenCV algorithms to calculate extrinsic camera positions and orientations automatically.
What output files does OpenCap generate?
OpenCap generates OpenSim-compatible .trc files containing 3D virtual marker coordinates and .mot files containing joint angles, moments, and ground reaction forces.
Can OpenCap calculate muscle forces without EMG?
Yes. OpenSim muscle-driven dynamic tracking simulations compute muscle-tendon forces and joint contact loads by optimizing muscle excitations to track kinematics while minimizing muscle activation costs.
What are the primary operational constraints of OpenCap?
OpenCap requires an active internet connection for cloud processing. Sub-optimal lighting, reflections, baggy clothing obscuring joint boundaries, or tripod movement during recording can degrade tracking accuracy.
Sources
- Uhlrich SD, Falisse A, Kidziński Ł, et al. “OpenCap: Human movement dynamics from smartphone videos.” PLOS Computational Biology, 19(10): e1011462, 2023. DOI: 10.1371/journal.pcbi.1011462. PMCID: PMC10586693. Read date: 2026-11-11.
- OpenCap Platform and Documentation: https://www.opencap.ai and https://docs.opencap.ai. Stanford University Neuromuscular Biomechanics Laboratory. Read date: 2026-11-11.
- Delp SL, et al. “OpenSim: open-source software to create and analyze dynamic simulations of movement.” IEEE Transactions on Biomedical Engineering, 54(11): 1940–1950, 2007. DOI: 10.1109/TBME.2007.901024. Read date: 2026-11-11.
- Lai AKM, Arnold AS, Wakeling JM. “Why are Antagonist Muscles Co-activated in Myocontrol? A Simulation Study.” Annals of Biomedical Engineering, 45(11): 2562–2574, 2017. DOI: 10.1007/s10439-017-1896-1. Read date: 2026-11-11.
- OpenCap Core Source Code: https://github.com/stanfordnmbl/opencap-core. Stanford University NMBL. Read date: 2026-11-11.
