Spotlight · 03

Every step.
Every phase.
Every stride.

ConductVision applies a Hidden Markov Model over per-foot contact and joint kinematics to produce smooth, gap-free temporal segmentation — turning a single sagittal video into stride-locked gait data.

  • Bilateral heel-strike / toe-off event detection
  • Cycle-aware smoothing eliminates flicker frames
  • Stride length, cadence, double-support, symmetry — one model
  • Exports CSV, C3D and clinical-report-ready summaries
Schedule a consultationHuman gait overview
gait-decoder.ipynb
LIVE · 30 FPS
REC
frame
2,148
time
00:01:11.6
stride
#42 of 87
phase
R-mid-stance
HMM segmentation
Foot phase & knee flexion decoder
Heel-strike
Stance
Pre-swing
Swing
L-FOOT
R-FOOT
L-KNEE
0–62°
R-KNEE
0–62°
0.0s
0.6s
1.2s
1.8s
2.4s
Stride length
1.42m
Cadence
112spm
Double support
21%
Symmetry
0.97

HMM · 4-state

Bilateral segmentation

30 FPS

Real-time decode, CPU-only

96.8%

Foot-phase F1 vs. force plate

Markerless

Single sagittal camera

Workflow

From a single video to stride-locked gait data

1

Record standard video

Use a single sagittal camera in a hallway, lab walkway, rehabilitation setup, or controlled motion protocol — no force plate or wearable sensors.

2

Decode with the HMM

A 4-state Hidden Markov Model reads per-foot contact and joint kinematics frame by frame, then smooths across the temporal sequence for gap-free stride boundaries.

3

Export reproducible metrics

Stride length, cadence, double-support, and symmetry export to CSV, C3D, and clinical-report-ready summaries you can take straight into statistics.

Applications

Metrics built for clinical and translational gait

The same HMM decoder powers human mobility studies and sits beside rodent gait and animal-behavior analysis, keeping a common video and export workflow across translational programs.

Gait speed
Step & stride length
Cadence
Stance & swing timing
Double-support time
Limb symmetry
Stroke rehabilitation research
Aging and fall-risk studies
Orthopedic mobility assessment
Translational rodent-to-human gait programs

Gait analysis — methods & FAQ

Do I need markers, a treadmill, or a force plate?
No. ConductVision gait analysis is markerless and runs on standard video from a single sagittal camera. The Hidden Markov Model infers foot-phase events and joint kinematics directly from pose landmarks, so no reflective markers, wearable sensors, instrumented treadmill, or force plate are required.
How accurate is the foot-phase segmentation?
In internal validation the 4-state HMM reaches 96.8% foot-phase F1 against force-plate ground truth. Cycle-aware smoothing eliminates the flicker frames that rule-based event detectors produce, giving clinically clean heel-strike and toe-off boundaries.
What metrics and exports do I get?
Per-stride and session-level outputs include stride length, cadence, double-support time, stance/swing timing, and bilateral symmetry. Results, keypoint coordinates, confidence values, and overlays export to CSV, C3D, and clinical-report-ready summaries for downstream statistics.
How does this relate to the Human Gait page?
The Human Gait page covers the broader markerless human-mobility workflow. This page focuses on the HMM stride-segmentation engine that powers it — the temporal decoding of foot phase and joint angles that turns raw pose tracks into stride-locked gait data.

Bring HMM gait analysis into your ConductVision workflow

Send a sample video or schedule a consultation to map the stride metrics your protocol needs.

Schedule a ConsultationExplore ConductVision