Open footage
Load a browser-compatible video or the included sample. No video is uploaded to PixelOkay.
Follow an object across a clip. Refine its trajectory, inspect confidence, and export coordinates or an annotated video—all from your browser.
MP4, WebM, MOV if supported by your browser · up to 200 MB · processed locally
Drag a box around an object, then track its motion.
Open a video or try the included moving-object clip.
Sampled coordinates and manual correction points.
Pause and draw a new tracking box, or use the selected object's coordinate inputs at the current time. Mark a frame lost during occlusion, then draw a visible correction when the object reappears. Re-run tracking to propagate corrections.
Coordinates in original video pixels.
Load a browser-compatible video or the included sample. No video is uploaded to PixelOkay.
Pause on an object and draw its box. Run patch matching, inspect confidence and add correction points if it drifts.
Save raw pixel coordinates, portable project JSON, transparent trails, annotated stills or a silent WebM clip.
The offline tracker uses visual template matching; it is not a trained neural tracker and cannot reliably maintain identity through occlusion, scale changes or camera cuts. Optional AI object detection downloads a public MediaPipe model from an external host only after you explicitly choose Download & Load in Model Setup, then runs on-device. Video remains local. Video export is silent and plays in real time; MP4 and fully automatic identity re-identification are not included.
Reading frames…
The application JavaScript did not finish initializing. Your browser work has not been cleared. Install the complete tool folder, then reload.
Track up to six manually selected objects through sampled video frames and export their positions as motion paths. Optional object detection can propose starting boxes on one frame; template tracking and manual review create the path.
Example: Track a cyclist across an unobstructed eight-second clip at 5 samples per second, review the marker where the rider passes behind a sign, add a correction and export the CSV path for another workflow.
Limits and privacy: Tracking supports up to six objects, 180 seconds and 1,800 sampled intervals; gaps marked lost are not measured positions. Template matching does not re-identify an object after long occlusion. Optional MediaPipe setup downloads code, WASM and model assets from external hosts; detection proposals use a selected frame, while video processing stays in the browser. WebM output is silent and browser-dependent.
No. The detector proposes boxes from a selected frame. Template matching follows the reviewed boxes across sampled frames and does not provide identity re-identification.
Yes. CSV coordinates use the source video’s pixel dimensions. Lost samples and gaps are marked or excluded rather than reported as observed positions.