Drag to move · pinch corner to resize · tap Done in Settings
Camera is off
Detection, tracking and counting all run right here in your browser. Nothing is uploaded.
Loading AI model…
0
People now
0
Entered
0
Exited
0
Total tracked
0Active
0Standing
0Sitting
0Walking
0Running
0Alerts
Made with Endless Love ❤️ — Designed & Developed by SAHIN
Settings
Performance
Mode
Keypoint confidence 0.30
Detection
Pose confidence threshold 0.30
Detection rate
Inference resolution
Tracking sensitivity (match distance) 0.22
Trail length
Counting
Counting line position 50%
Line orientation
"Entry" direction
Count cooldown per track 1000ms
Zones
Zone A
Show and count a detection zone
Display
Bounding boxes
Tracking ID
Confidence %
Movement trail
Direction arrow
Activity status (standing/sitting/walking/running/jumping/bending/lying/falling)
Pose skeleton
Tracking duration
Zones overlay
Performance HUD (FPS/latency)
Privacy & Data
Cloud analytics
Sends only aggregate counts — never video — to your Worker
Video frames are analyzed with an on-device model and are never uploaded, recorded, or sent to any server — with or without analytics enabled.
Session
About
AI People Tracker v2.0.0 — on-device multi-person pose estimation (TensorFlow.js MoveNet MultiPose, up to 6 people at once) with a full 17-point skeleton, geometry-based activity recognition (standing, sitting, walking, running, jumping, bending, lying down, falling), and time-aware identity tracking that keeps boxes smooth and re-associates a person's ID after a brief occlusion or exit from frame using position and motion — this is a lightweight, best-effort heuristic, not facial recognition or biometric identification.
Activity classification is derived purely from body-joint geometry and motion, not a trained action-recognition model — treat "Falling" alerts as a heads-up, not a certified detection.
Requires HTTPS and a browser with camera + WebGL support.