Live Analysis Manual

1 Opening the page

• Sign in with your Google account or email link — authentication is required before you can access the analysis page.
• This is a password-free account — when you enter an email address, you will receive an email with a confirmation link to log in. Google authentication is automatic.
• A model-loading overlay appears with a progress bar while the pose detection model downloads and initialises.
• On desktop, the Heavy model loads by default; on mobile, the Lite model is used instead. After the first load, the model is cached so subsequent visits start faster.
• Once the model is ready the overlay disappears and you see the camera selection area.

2 Starting a camera

• The Camera dropdown lists all available video input devices on your system (built-in webcam, USB cameras, virtual cameras, etc.).
• Select a camera and click Start Camera.
• Your browser will prompt for camera permission the first time — grant it to proceed.
• Once started, the live video feed appears on the main canvas and real-time pose inference begins immediately.
• The Start Camera button is replaced by the Add Camera button for multi-camera setups.

3 Adding cameras

• After starting the first camera, click Add Camera to add additional cameras.
• Select a different camera from the dropdown and click Add Camera — you can add as many as your hardware supports.
• A thumbnail strip appears above the main canvas showing each active camera feed.
• Click a thumbnail to switch which camera is displayed as the primary (large) view.
• Each camera runs its own independent pose inference loop.
• The Active cameras list in the sidebar shows all cameras with their labels.

4 Camera settings

• Expand the Camera Settings accordion (appears after starting a camera) to adjust resolution, frame rate, and other hardware-level settings for the active camera.
• Available settings depend on your camera hardware and browser support.
• Expand the Lens Intrinsics accordion to enter focal length, principal point, and distortion coefficients if you have calibrated your camera externally.

5 Session details

• Open the Session details accordion in the sidebar to fill in metadata about your analysis.
• Title (required) — give your session a descriptive name.
• Subject alias (required) — identify the person being analysed. Avoid using full names for privacy.
• Notes (optional) — free-text area for clinical observations, setup descriptions, or any relevant context.
• Keywords (optional) — comma-separated tags for categorisation (e.g. "knee pain, gait, trendelenburg").

6 Calibrating

• Open the Calibration accordion in the sidebar.
• Enter a known real-world measurement (e.g. height of the subject, length of a known object in the frame).
• Select the unit: Meters, Yards, Foot, or Pixels.
• Click Draw reference and draw a line on the canvas matching the known measurement.
• Toggle Show reference line to keep the calibration line visible.
• Set the Video speed if your camera feed runs at a non-standard rate.
• Set the Data sampling rate (FPS) to control how frequently pose data is sampled.
• Click Apply calibration to activate real-world units. Until calibration is applied, all measurements are in pixels.

7 Visual options

• Open the Visual options accordion to control what is drawn on the canvas.
• Toggle switches for: Chronometer, Keypoints, Skeleton, Angles, 3D angles (experimental), Center of mass, Inertia vector, and Anonymization.
• When Angles is enabled, expand the angle-reversal sub-options to reverse the measurement direction for individual joints (R/L Elbow, Knee, Shoulder, Hip, Ankle).
• Adjust the Score threshold slider (0.00–1.00) to filter out low-confidence keypoints.
• Click Body parts to expand a panel where you can toggle visibility for specific body regions: Face & head, Upper limb, Hands, Lower limb, Feet. Use the All / None buttons for quick selection.

8 Models

• Open the Models accordion to configure pose and hand detection models.
• Pose Model: toggle Pose tracking on/off. Select the model variant: Lite (fastest), Full, or Heavy (most accurate). Toggle Smoothing to reduce jitter.
• Hand Model: toggle Hand tracking to enable hand landmark detection. Hand tracking is marked as Preview, indicating it is an experimental feature. Select Lite or Full model. Toggle Anchor hands to body to connect hand landmarks to the pose skeleton.
• A spinner appears briefly while a new model loads.

9 Recording

• Click the record button () in the transport bar at the bottom of the screen to start capturing.
• An elapsed timer in red appears showing recording duration.
• While recording, pose data (keypoints, angles, velocities, etc.) is captured alongside the video frames.
• Click the record button again to stop. The recording is added to the Recordings list in the sidebar.
• You can make multiple recordings within the same session.

10 Reviewing recordings

• Click on a recording in the Recordings list to enter review mode.
• The transport bar updates with a timeline seekbar so you can scrub through the recorded footage.
• Use the frame step buttons ( / ) to move one frame backward or forward.
• Use the speed selector () to adjust playback speed: 100%, 75%, 66%, 50%, 30%, or 10%.
• Charts below the video update to show the data from the selected recording.
• Each chart has a notes textarea below it — use these to record observations about the data.
• Click Back to Live to return to the live camera feed.

11 Annotations

• The annotation toolbar appears below the main video canvas.
• Use the drawing tools to annotate directly on the video frame: Select, Free draw, Text, Rectangle, Circle, Line, Arrow, Protractor, and Ruler.
• Use the colour picker to change the stroke colour.
• Click Undo () to remove the last annotation, or Clear all () to remove all annotations.
• Toggle annotation visibility with the eye button ().
• Annotations are stored per recording and saved alongside the session.

12 Saving

• Click Save and download in the bottom-left corner to open the save menu.
• Choose Publish private session (paid tiers) or Publish public session (all tiers) to save the session to your account.
• A save confirmation modal appears — fill in required session details (title, subject alias) if not already provided.
• Sessions are auto-drafted periodically so work is not lost if you close the browser.
• Saved sessions can be reopened from My Sessions ( /history/).

13 Exporting

• Open the Save and download menu and click Export data.
• The export dialog offers the following options:
  • Scope: export the Current take or All takes (bundled as a ZIP file).
  • Format: CSV (spreadsheet-ready) or JSON (structured data).
  • Layout (CSV only): Wide (Excel friendly) or Tall/tidy (Looker Studio).
  • Timestamps (CSV only): Quantized (grid-snapped) or Raw (unquantized).
  • Data columns (CSV wide only): select which columns to include from a set of checkboxes.
• Exported data includes all captured keypoints, joint angles, velocities, accelerations, and centre-of-mass positions for each frame.

Camera Setup

• The camera dropdown enumerates all videoinput devices reported by the browser's MediaDevices API.
• After selecting a camera, click Start Camera. The browser requests permission, and once granted, the feed renders on the main canvas.
• The canvas automatically resizes to match the camera's native resolution.
• If a camera is disconnected or access is revoked, an error message is shown.
• Use Reset analysis to start over to stop all cameras and clear the session.

Multi-Camera

• After the first camera is started, click Add Camera to attach additional cameras.
• A thumbnail strip appears above the main canvas showing all active camera feeds.
• Click any thumbnail to switch the primary (large) view to that camera.
• Each camera runs its own independent inference loop — pose data is tracked separately per camera.
• The Active cameras list in the sidebar shows a labelled entry for each running camera.
• When using multiple cameras and reviewing a recording, charts display data from the selected camera. Use the chart camera selector dropdown above the charts area to switch which camera's data is displayed.

Pose Models

The pose landmarker provides three model variants. Switch between them in the Models accordion.

• Model files are cached after the first download, so subsequent loads are fast.
• You can switch models at any time — a spinner indicates loading.
• Toggle Pose tracking off to disable pose detection entirely (e.g. when using hand tracking only).

Hand Tracking Preview

• Enable Hand tracking in the Models accordion to detect 21 hand landmarks per hand.
• Choose Lite (faster) or Full (more accurate) hand model.
• Anchor hands to body connects hand landmarks to the nearest wrist keypoint from the pose model, creating a unified skeleton.
• Hand tracking runs in parallel with pose detection but adds processing overhead — consider using a lighter pose model if frame rate drops.

Smoothing

• Toggle Add smoothing to remove jitter from video in the Models accordion.
• When enabled, temporal smoothing is applied to landmark positions, reducing frame-to-frame jitter at the cost of slightly delayed tracking.
• Smoothing is enabled by default and is recommended for most use cases.
• Disable smoothing for fast-moving activities where low latency is more important than smooth tracking.

Calibration

• Spatial calibration: enter a known reference measurement, select a unit (Meters, Yards, Foot, Pixels), draw a reference line on the canvas, and click Apply calibration. All distance-based measurements (displacement, velocity, acceleration) will be converted to real-world units.
• Video speed: set the playback speed multiplier if your camera feed or recording was captured at a non-standard rate (e.g. slow-motion).
• Data sampling rate: choose how many frames per second to sample for analysis data. Options range from 23.976 FPS up to Limitless.
• Until calibration is applied, all measurements default to pixels. A warning indicator is shown: "Units in px until calibration."
• Toggle Show reference line to keep the drawn calibration line visible on the canvas.

Visual Overlays

• Score threshold (0.00–1.00): keypoints below this confidence score are hidden. Default: 0.30.
• Body parts: granular control over which body regions are displayed — Face & head, Upper limb (shoulder to wrist), Hands (pinky, index, thumb), Lower limb (hip to ankle), Feet (heel, foot index).
• Angle reversal: when Angles are enabled, you can individually reverse the measurement direction for R/L Elbow, Knee, Shoulder, Hip, and Ankle joints.
• Mirror: click the mirror button () in the transport bar to horizontally flip the video. Useful when the camera faces a mirror or when you want the display to match the subject's perspective.

Recording & Playback

• Record: click the record button () in the transport bar to start/stop recording. While recording, a red elapsed timer is displayed.
• Recordings list: all recordings appear in the Recordings sidebar accordion. Click a recording to enter review mode.
• Review mode: in review mode, the live feed is replaced by the recorded video. A seekbar appears in the transport bar for scrubbing.
• Frame step: use (previous frame) and (next frame) for precise frame-by-frame navigation.
• Playback speed: click the speedometer icon () to choose from 100%, 75%, 66%, 50%, 30%, or 10% playback speed.
• Snapshot: click the camera icon () to capture a still image of the current canvas frame (including overlays).
• Undo/Redo: use the undo () and redo () buttons or keyboard shortcuts (Ctrl+Z / Ctrl+Shift+Z).
• Back to Live: click Back to Live in the Recordings accordion to return to the live camera feed.

Charts

Charts are rendered below the video canvas and update with data from the selected recording.

• Each chart has a notes area below it for text annotations.
• The chart toolbar allows zooming, panning, and downloading chart images.
• In multi-camera setups, a chart camera selector dropdown lets you switch which camera's data is plotted.

3D Skeleton Experimental

• A 3D canvas renders a real-time 3D skeleton based on the pose model's world landmarks.
• Open the 3D Rendering Options accordion to configure:
  • Enable 3D canvas — toggle the 3D view on/off
  • Enable zoom — allow mouse-wheel zoom
  • Zoom speed — adjust zoom sensitivity
  • Auto-rotate — slowly rotate the camera around the skeleton
  • Dark background — switch between light and dark canvas backgrounds
  • Show ground grid — display a reference grid at ground level
  • Reset camera — return the 3D camera to its default position
• Click and drag to orbit the 3D camera. Right-click and drag to pan.

Annotations

The annotation toolbar provides drawing tools to mark up the video frame. Annotations are rendered on an overlay canvas.

• Use the colour picker to change the stroke colour of new annotations.
• Undo removes the last annotation. Clear all removes every annotation.
• Toggle visibility with the eye button to hide/show all annotations without deleting them.
• Protractor displays the measured angle in degrees. Ruler displays the measured distance in the current calibration unit.

AI Chat Assistant

• Click Chat assistant in the top nav bar or below the video to open the chat panel.
• The assistant can help with camera setup, assessment type selection, data interpretation, and more.
• Starter prompts are provided: introduction/walkthrough, gait analysis recording tips, and chart annotation tutorials.
• Type a question and click send. Use the trash button to reset the conversation.
• Free and Explorer tiers use a basic chat assistant. Premium tiers unlock advanced chat capabilities.

Saving a Session

• Click Save and download to access save options.
• Publish private session: available on paid tiers — saved sessions are only visible to you.
• Publish public session: available on all tiers — saved sessions are publicly visible.
• Before publishing, ensure Session title and Subject alias are filled in — they are required.
• The session data includes all recordings, annotations, calibration settings, chart notes, and metadata.
• Sessions are auto-drafted periodically, so in-progress work is preserved if the browser is closed.
• After saving, a confirmation appears with a link to the saved session.

Data Export

Export captured data for external analysis in your preferred tool.

Export Dialog Options

Data Columns (CSV wide layout only)

When using CSV wide layout, choose which data columns to include. Columns enabled by default are marked below.

• Exported data includes: frame timestamps, all 33 pose landmarks (x, y, z, visibility), computed joint angles, velocities, accelerations, and centre-of-mass coordinates.
• Access the export dialog from the Save and download menu by clicking Export data.