See What We Count Without Missing Anything¶
AITracker lets you inspect and analyze traffic measurements, including traffic belonging to a selected movement or vehicle category.
Vehicle tiles¶
The Gallery tab displays vehicles from the currently loaded location in a scrollable image grid. It uses the images referenced by classification results and supports the older Images/<TUID>Tracker.jpeg format. A missing or damaged image is replaced with a placeholder. The first row of each tile contains the TUID, event time, and counting-status marker. The second row contains up to two movements and two classes, displayed only as coloured icons. Recognized registration text is overlaid on the image when available.
Filters for text, vehicle class, movement, source file, and image availability are in the left panel. Class options use their corresponding icons; movement options show the configured colour and direction arrow. Select a tile with the mouse or navigate the grid with the arrow keys. A larger preview and object details then appear on the right. The details show the object image and its trajectory on the camera background together with the configured gates. Vehicle classes use icons, while movements use a name, colour, and direction arrow.
Click the object image or trajectory preview to open it over the whole window. The image animates from the clicked preview and returns to it when closed. Close it with the × button, Esc, or a click on the dark background. The trajectory preview uses the framing configured in Design so its background, gates, and vehicle path share one coordinate system. If the correct background variant is missing, the preview does not substitute a background assigned to another recording.
The class and movement lists let you enter corrections without opening the Results table. Each class icon includes the algorithm confidence percentage, and the options are sorted from the highest score. A changed object receives a yellow tile, and the correction uses the same persistence mechanism as the Results table. Details also show the source filename, event date and exact time, classification confidence, exact validation state, speed, passage time, registration text, and plate crop when available. Speed uses green, yellow, orange, and red backgrounds at thresholds of 40, 90, and 140 km/h. The TUID and time link opens the correct recording at the object's frame in Results. The browser does not create or edit user tags.
Video window — results in motion¶
AITracker GUI displays the current video in an internal window. Every frame and all object-drawing layers enabled by the user are displayed. Toggle these layers with the buttons on the right.
Charts¶
Result charts appear below the video and use the same time grouping selected in Design, which is five minutes by default.
Chart types include:
- vehicle count by movement,
- vehicle count by section; only one section exists by default,
- vehicle count by category in the selected vehicle classification,
- a speed scatter plot grouped into five-minute and 10 km/h intervals.
Video and time slider on a chart¶
Each chart has a red area showing the duration of the current video and a red vertical line showing the displayed moment. A white indicator attached to the line displays the current measurement time.
In point mode, the tracker label always keeps its TUID readable. An uncounted tracker has a separate red X box beside the identifier; its movement and category remain visible to help diagnose why it was rejected. The same large red X starts the information row on an uncounted object's Gallery tile.
Results table¶
The results table is located in the Results tab. It contains all vehicles appearing in the displayed video, including vehicles crossing the boundary between two recordings. It can be sorted by every column.
Mouse actions:
- Left-click a row to jump to that tracker or object.
- Right-click a row to open a context menu with additional actions.
- Hover over a vehicle's Category cell to display the tracker image selected during processing.
DAV format limitations¶
- For DAV files, Jump to Vehicle may be slower because seeking through the recording is often required.
- This behaviour results from the format and decoding process and does not necessarily indicate a classifier error.