Posture intelligence,
running privately
in your browser.
PostureSense uses browser-native computer vision and MediaPipe WebAssembly to analyze body movement in real time without sending your camera stream to the server.
How on-device detection works
Your camera feed stays in your browser. Every video frame moves through a 4-step local analysis loop.
Camera Capture
HTML5 MediaDevices API accesses your webcam locally. No video bytes leave your device.
MediaDevices.getUserMedia()
WASM Inference
MediaPipe Tasks Vision WebAssembly extracts 33 3D body keypoints off the main thread in a WebWorker.
poseLandmarker.detect()
Biomechanics
EMA filtering smoothes spatial jitter while joint angles, symmetry, balance, and stability are calculated.
calculateBiomechanics()
Rule & Cues
Measured joint angles are matched against configured rules in yoga_poses.json to output real-time cues.
evaluatePoseRules()
What PostureSense measures
Every posture session evaluates 8 core movement dimensions in real time using objective geometric calculations.
Form Score (0–100%)
Evaluates how accurately your joint positions match the target angle constraints defined in the pose configuration.
Biomechanical Symmetry
Compares left and right limb extension, shoulder height levelness, and lateral body alignment.
Center of Mass Balance
Measures body mass distribution over your support base to detect lean and sway during posture holds.
Form Stability
Quantifies spatial steadiness over time, detecting subtle trembling or postural instability.
Range of Motion (ROM)
Calculates exact joint articulation angles (knees, hips, shoulders, spine) during movement transitions.
Tracking Quality
Monitors keypoint visibility confidence so you know when lighting or distance affects camera detection.
Sustained Hold Time
Tracks the continuous duration a recognized static pose is held with correct form.
Repetitions & Tempo
Counts completed exercise reps using state machine transition gates and range-of-motion depth thresholds.
Poses recognized by PostureSense
Evaluates front knee bend (80–110°), rear knee lock (160–180°), and horizontal arm extension.
Baseline posture alignment testing horizontal elbow extension (165–195°) and shoulder levelness.
Measures single-leg standing balance (160–180° knee) and lifted knee outward rotation (30–90°).
Evaluates thoracic spine extension (10–45° arch) and grounded hip alignment (150–180°).
Real-time guidance as you practice
PostureSense provides dynamic visual feedback on screen during your practice session without interrupting your movement.
Pose & Confidence Meter
Instantly shows which pose is detected along with real-time classification confidence percentage.
Form Correction Badges
Highlights specific form violations (e.g., "bend front knee further", "level shoulders") as rule checks fail.
Hold & Rep Tracking
Tracks continuous hold duration for static poses and counts completed repetitions for dynamic exercises.
Skeletal Landmark Canvas
Renders high-DPI keypoint overlays colored by tracking confidence directly over your live video preview.
What happens after your session
Every completed session builds your personal posture intelligence dataset.
Calculates linear regression trends over 7-day, 30-day, or all-time practice periods.
Automatically tracks best score, longest hold time, top symmetry, stability, and ROM.
Surfaces action items based on form rule history without black-box AI hallucinations.
Comprehensive reporting & exports
Export your movement data for personal tracking, coaching reviews, or health records.
Session Report
Single-session breakdown featuring overall form score, joint angle summary, biomechanics radar, and failed rules.
Progress Report
Longitudinal progress evaluation analyzing practice frequency, average score movement, and personal records.
Pose / Exercise Report
Pose-specific history compiling all practice attempts, best form scores, and hold duration records for a single pose.
Comprehensive Report
Portfolio executive overview summarizing overall progress, score trends, biomechanics trends, and session logs.
Multi-Format Authenticated Exports
Download your reports on-demand in PDF, RFC-8259 JSON (schema v2.0.0), or RFC-4180 CSV formats.
100% On-Device Perception
Local WebAssembly Execution
PostureSense performs 100% of computer vision pose detection inside your web browser using MediaPipe WebAssembly. Your webcam feed never leaves your device.
- Zero Video TransmissionNo video frames or raw webcam images are ever uploaded to cloud servers.
- Zero Raw Landmark PersistenceRaw 33-point keypoint streams are processed transiently in memory and discarded.
- Telemetry-Only SavingOnly finalized session metrics (scores, duration, rep counts) are saved upon explicit user save.
Authenticated User Data Isolation
All persisted session summaries, analytics, and report downloads are protected with strict user isolation guards (IDOR protection).
Your practice history remains strictly private to your authenticated account, protected by encrypted session cookies and strict CORS policies.
Built with engineering, vision & curiosity.
PostureSense was designed and developed as a browser-native computer vision platform, combining real-time perception, biomechanics, analytics, and production engineering.
Designed & Engineered by Vedaang Sharma · Jaipur, Rajasthan, India
Vedaang Sharma
Lead Developer & ArchitectDesigned and engineered PostureSense v2 across its browser-native perception pipeline, biomechanics analysis, rule-based pose recognition, scoring, analytics, dashboard, reporting, and production architecture.
Frequently Asked Questions
Instant answers to common questions about on-device AI perception, privacy, and biomechanical tracking.
What is PostureSense and how does it work?
PostureSense is a browser-native AI movement intelligence platform. It runs Google MediaPipe Tasks Vision WebAssembly directly inside your client browser to detect 33 3D skeletal landmarks at 30–60 FPS. By calculating vector angles between anatomical joints in real time, it compares your movements to biomechanical constraints and delivers instant visual feedback without sending video data to external servers.
Does PostureSense record, store, or stream my webcam feed?
No. PostureSense employs a 100% on-device, client-side perception architecture. All video frames from your camera are processed strictly inside local WebWorker memory buffers and are immediately discarded. No video bytes, images, or raw coordinate streams are ever uploaded or transmitted across the internet.
How does 33-point landmark tracking and real-time biomechanics calculation work?
The pipeline uses MediaPipe 3D PoseLandmarker combined with an Exponential Moving Average (EMA) spatial filter to remove keypoint jitter. Joint articulation angles (such as knee flexion, hip grounding, elbow lock, and thoracic spine extension) are calculated using 3D vector dot products and compared against deterministic rule bounds defined in yoga_poses.json.
Which yoga poses and posture exercises are currently recognized?
PostureSense currently features configured rules for Warrior II (Virabhadrasana II), T Pose (Calibration Baseline), Tree Pose (Vrikshasana), and Cobra Pose (Bhujangasana). Each pose evaluates specific target joint angles, symmetry bounds, and stability thresholds.
What 8 biomechanical metrics are evaluated during practice?
Every session tracks 8 objective metrics: (1) Form Accuracy Score (0–100%), (2) Bilateral Symmetry, (3) Center of Mass Balance, (4) Form Stability, (5) Joint Range of Motion (ROM), (6) Camera Tracking Quality, (7) Sustained Hold Time, and (8) Repetition/Tempo counts.
Is PostureSense free to use? Do I need special hardware?
PostureSense is 100% free and open-access. No special hardware, sensors, wearable devices, or app downloads are needed—just a standard webcam and any modern web browser (Google Chrome, Microsoft Edge, Mozilla Firefox, or Apple Safari).
Stop guessing your form.
Start measuring it.
Experience real-time computer vision pose analysis locally in your browser.
Get in touch
Have questions or feedback? We’d love to hear from you.