Biomechanical assessment of patient movement during rehabilitation using an optical motion capture system — from markers to clinical decision
The accuracy of the entire subsequent biomechanical analysis is determined by the quality of camera calibration and marker placement reproducibility. Optical systems based on passive light-reflective markers (Vicon, Qualisys) or modern markerless solutions (Theia3D, OpenCap) reconstruct the position of anatomical references in space with submillimeter accuracy at a frequency of 100–200 Hz — sufficient to capture even the fast phases of foot strike and finger push-off.
A classical optical system (Vicon Vantage, Qualisys Miqus) consists of 8–12 infrared cameras arranged around a circle in the recording area. Each camera emits an IR flash that reflects off spherical markers covered with retroreflective material (diameter 9-14 mm). Each camera captures a 2D projection of the marker; triangulation from at least two angles gives a 3D coordinate with an error <1 mm.
Markerless systems (Theia3D, OpenCap, DeepLabCut-based pipelines) use convolutional neural networks to directly detect anatomical keypoints on video from ordinary cameras without physical markers on the body. This speeds up patient preparation (no need to apply 39 markers) and reduces error associated with soft tissue displacement relative to bone, but marker systems still outperform in terms of accuracy for rotational components of motion.
The choice of system in rehabilitation practice depends on the task: a marker system is the gold standard for scientific and surgical solutions (e.g., ACL revision), while markerless systems are convenient screening tools for weekly monitoring of progress.
Static volume calibration is performed with a stationary L-like frame with known distances between markers; dynamic calibration uses a T-like pole that the operator describes in space over 60-90 seconds. The system calculates internal and external camera parameters using bundle adjustment.
A modified Plug-in Gait model (Vicon Nexus) is the most common protocol for clinical gait analysis. Markers are applied to anatomical landmarks:
• Torso and pelvis: acromion processes, jugular notch, C7, T10, anterior and posterior superior iliac spines (ASIS/PSIS) • Thigh: lateral femoral condyle + auxiliary 'wand' marker for knee axis determination • Shank: lateral malleolus + auxiliary shank marker • Foot: heel, head of the second metatarsal
The accuracy of repeated marker placement (test-retest) depends on the operator's experience: the error in determining the knee joint axis between sessions typically ranges from 2–4°, which is the main source of variability in clinical angular data and why longitudinal observation of a single patient is entrusted to the same specialist.
Before analyzing movement, the system must 'understand' the geometry of the specific body: segment lengths, joint axis positions, mass distribution. The static test — a short (2–3 s) recording of the patient in neutral position — provides these parameters and transforms the cloud of 3D marker points into a biomechanically interpreted skeletal model with defined joint centers.
The software calculates from the coordinates of static markers:
• Segment lengths (thigh, shank, foot) — basic measurements for scaling the dynamic model • Width of the knee and ankle joints (for calculating the joint axis) • Position of the hip joint center — regression equations by Devis/Bell based on pelvis width, OR the functional method SCoRE (Symmetrical Center of Rotation Estimation), which determines the center of rotation by analyzing thigh movement relative to the pelvis during passive circular movements of the leg
The functional method is more accurate (error ~6-8 mm vs. 15-20 mm in regression formulas), but it requires an additional trial with active movement of the distal end, which may not be feasible in patients with limited mobility after surgery.
Each bony segment is assigned a local orthogonal coordinate system (ISB standard) constructed from three or more markers. This allows describing the relative joint motion through Euler/Cardan angles (e.g., sequence Grood-Suntay for knee: flexion-extension → abduction-adduction → internal-external rotation).
The skeletal model forms a kinematic chain: pelvis → thigh → shank → foot on one side, pelvis → torso → shoulder → forearm → hand on the other. The motion of each distal segment is described relative to the proximal one, producing clinically interpretable "joint angles" instead of absolute marker trajectories in space.
The patient walks the measurement track several times at a natural pace. The system simultaneously records the 3D trajectories of all 39 markers, the vertical and horizontal components of ground reaction force from force platforms embedded in the floor, and, when clinically indicated, surface electromyography (EMG) of the quadriceps femoris, gastrocnemius, and other key muscle groups.
During gait, markers periodically 'disappear' from camera view — replaced by another end, clothing, or return to the side of the camera. Software tracks each marker's identity (labeling) and applies gap-filling algorithms: spline interpolation for gaps <10 frames, or kinematic reconstruction based on segment rigidity (rigid body fill) for longer gaps when the motion of adjacent markers of the same segment is known.
After filling, the trajectory is filtered with a low-pass Butterworth filter (4th order, cutoff frequency 6 Hz), which removes high-frequency tracking noise while preserving physiologically plausible joint dynamics.
One walking trial of 10 m at 1.2 m/s lasts about 8 seconds and generates over 1200 frames × 39 markers × 3 coordinates — around 140,000 numerical values processed in a few seconds.
Force plates measure three orthogonal components of force (vertical, anterior-posterior, medial-lateral) and moment, allowing the center of pressure (center of pressure) under the foot during the stance phase to be calculated. The vertical GRF during normal walking has a characteristic M-like shape with two peaks (~110–130% body mass) at the beginning and end of the stance phase and a trough (~70–80%) in the middle of support.
In parallel, the system automatically determines the spatial-temporal gait parameters from the moments of heel contact and toe-off:
• Cadence (strides/minute) — walking pace • Length of step and length of double step (stride length) • Stance phase / swing phase / double stance phase time • Width of support
These parameters are the first and most sensitive indicators of a pathological gait pattern that a clinician assesses before a detailed angular analysis.
The black 3D trajectories of markers are transformed into clinically interpreted graphs: flexion-extension (sagittal plane), abduction-adduction (frontal plane) and internal-external rotation (transverse plane) for each lower extremity joint, built relative to the normalized gait cycle from 0% (heel strike) to 100% (next heel strike of the same leg).
The angle between two adjacent segments (for example, the tibia and femur for the knee) is calculated not as a simple angle between vectors, but as a sequence of three independent rotations around moving axes (Cardan angles, Grood-Suntay convention, ISB standard):
1. Flexion-extension — rotation around the mediolateral axis of the tibia 2. Abduction-adduction — rotation around the floating (floating) axis, perpendicular to the two previous ones 3. Internal-external rotation — rotation around the longitudinal axis of the femur
This sequence is not commutative — the order of rotations is fundamentally important and clinically standardized to ensure comparability between laboratories and patients.
Each walking cycle (from heel strike to the next heel strike of the same foot) is normalized to a scale of 0–100%, then 5–8 recorded cycles are averaged with standard deviation calculation at each point — this forms the characteristic 'corridor curve' on the clinical report.
The sagittal plane (maximum range of motion) is sensitive to quadriceps weakness, pain, and contractures: reduced flexion of the knee in the swing phase ('stiff-knee gait') is typical for hemiparesis; reduced knee extension at the end of stance is a sign of a painful guarding pattern in osteoarthritis.
The frontal plane reveals compensatory mechanisms: excessive lateral trunk lean (Trendelenburg gait) due to weakness of the gluteus medius muscle, valgus knee loading — a risk factor for ACL re-tear.
The transverse plane (with the smallest amplitude, most sensitive to marker error) reflects rotational compensations of the tibia and femur, important in assessing torsional deformities in children and post-rotational osteotomies.
The classic double-hump curve of knee flexion — one of the most reliable markers of normality: the first peak (~15-20°) during loading phase cushions the impact, while the much larger second peak (~60-65°) in swing phase ensures clearance of the foot over the ground. Loss of either peak is a specific diagnostic sign.
The final step converts calculated kinematic curves into clinically applicable solutions. Patient data are overlaid on a normative corridor (mean ± 1 standard deviation for >500 healthy individuals of the corresponding age), the limb symmetry index (LSI) is calculated, and results are documented for tracking rehabilitation progress over time and making decisions about clearance to sport-specific loading.
Limb Symmetry Index is calculated as the ratio of the index of the injured (or operated) leg to the healthy one:
LSI (%) = (value of the injured leg / value of the healthy leg) × 100
Applied to peak moments of knee extension force, maximum flexion angle in swing phase, stride length, and stance time. LSI <85–90% after ACL reconstruction is associated with significantly higher risk of recurrent injury upon return to sport and serves as a basis for delaying clearance by the coach/physician.
The patient's curve is compared to the normative corridor using the Gait Deviation Index (GDI) or Gait Profile Score (GPS) — a generalized one-dimensional metric (standard deviations from the norm) that allows monitoring the overall ‘quality’ of walking with one number from session to session, without delving into all 9 angular curves each time.
In post-operative management (ACL reconstruction, total knee arthroplasty, osteotomy), repeated motion capture sessions are conducted at control points — 3, 6, 9, and 12 months — to objectively track recovery of ROM, symmetry of loading, and walking speed, not just based on subjective complaints.
The decision for return-to-play (return-to-sport) in the modern protocol is based not only on the time since surgery but on a comprehensive set of objective criteria: LSI strength and kinematics >90%, pain-free full ROM, absence of compensatory patterns in the frontal plane (valgus collapse), as well as psychological readiness (ACL-RSI questionnaire). Motion capture is the only tool that provides a quantitative, reproducible answer to the question ‘does the knee move just like a healthy one?’, rather than relying on subjective assessment from an examination.
Meta-analyses show that athletes allowed back into competition based solely on time after surgery (without objective RTP testing) have up to 4 times higher risk of ACL re-tear within the first 12 months compared to those who underwent quantitative biomechanical testing.
| Product | Indication | Trial Design | Key Result |
|---|---|---|---|
| Healthy control | Reference norm | Symmetrical double-hump pattern of knee flexion during gait, full ROM in all planes | LSI 96–99% |
| After ACL reconstruction | Operated leg | Quadriceps avoidance, reduced peak knee flexion in swing phase | LSI 78–86% |
| Hemiparesis (stroke) | Paretic leg | Stiff-knee gait, circumduction of the hip, reduced amplitude in all joints, slower cadence | LSI 52–68% |
| Knee osteoarthritis | Affected joint | Antalgic (pain-avoidance) pattern, incomplete extension at end of stance, shortened stride | LSI 80–88% |