Upload a video from a gate, parking lot, roadside or dashcam. Every plate is found frame by frame, followed as it moves, and read many times; the readings are combined into one row per vehicle — also when several cars are in view at once. Everything runs in your browser; the video never leaves your device.
Loading models…
Tip: walk slowly past parked cars and keep each plate in view for about 2 seconds, readable to your eye (3–10 m). Fast pans still work, but give fewer reads per car.
Choose a video to start.
Plate
Type
First seen
Last seen
Reads
Confidence
Route
Help train the model: press ✓ Correct when a plate is read right, or ✏️ Wrong? and type the correct text — each tap sends the frame to the tz-alpr team. Pressing it on a correct row and saving unchanged confirms the reading, which also helps.
Route: auto accept ≥ 90 % and the same text on most frames · review 70–89 % or a series not yet proven on real photos · manual below 70 %. A plate read on only one frame is listed only when that read is confident, and then always goes to review.