Dictionary

The glossary presents an alphabetical list of all terms used in the documentation that may be specific to AISP software and the AITracker system.

---

Technology Terms

Detection — A single, one-frame detection of an object by AI. Each frame of the video is analyzed by AI, resulting in detections that include the object's category and location. Detections are then combined into a Tracker object.

Digital ZOOM (Cropping) — A feature that digitally enlarges a selected area of the video before analysis. This improves AI detection accuracy for small or distant objects.

Fréchet Distance — A mathematical metric used to calculate the similarity between a vehicle's movement trajectory and a user-drawn relation curve. Used for assigning vehicles to specific traffic relations.

Tracker — A single object tracked across multiple video frames. A tracker is created by combining detections from consecutive frames, tracking the object's movement. Each tracker receives a unique identifier called TUID (Tracker Unique Identifier) and can be in a valid or invalid state. It has its own trajectory and characteristics used for classification.

TUID (Tracker Unique Identifier) — A unique identifier assigned to each tracked object (tracker) in the system. Used for referencing specific vehicles or pedestrians in results and errata.

VRT (Vehicles Realtime Tracking) — AISP's proprietary object tracking algorithm (MOT - Multiple Object Tracking) that ensures high accuracy in identifying and maintaining vehicle identity throughout a video recording.

YOLO (You Only Look Once) — A breakthrough neural network architecture used by the system for simultaneous localization and classification of multiple objects in real-time. Enables fast and accurate detection of vehicles, pedestrians, and other road users.

Configuration Elements

Cross-section — A measurement line or curve drawn by the user to filter and categorize results. Objects are only counted in spreadsheets after crossing this line. Used to define specific measurement points on the road.

RAW Categories — The internal classification of vehicles and pedestrians (22 classes) recognized by AI. These are then mapped to target classifications (e.g., GPR, Euro-6, SCPR) based on user configuration.

Relations — User-drawn movement curves that form the basis for traffic intensity measurements. The system assigns vehicles to specific directions (e.g., straight, left, right) based on how closely their trajectory matches these curves using Fréchet Distance.

Data Processing Functions

Errata — A tool for manually correcting results (e.g., changing category or relation) without modifying the original AI-processed data. Changes are saved in an errata.pickle file and can be reverted at any time.

Reclassification — The process of recalculating and reclassifying already-counted objects (trackers) after configuration changes (e.g., adding new relations, changing cross-sections). Performed locally without re-uploading video to the cloud.

Traffic Designer (Projektant Ruchu) — A built-in tool for automatically creating traffic flow matrices and analyzing transit patterns by matching vehicles from multiple measurement points based on license plate recognition (ALPR).

Validated — A status indicating whether a given object (tracker) is included in the final results spreadsheet. Only validated objects appear in reports and statistics.

Tracker Validation Statuses

Each tracker can have one of the following validation statuses:

• Unknown — Initial state after creation, before the validation process is completed.

• Validated — Tracker has been properly validated and will be counted in the final results.

• AreaUncounted — The counting area is disabled or the tracker did not enter the designated measurement area.

• RelationUncounted — The tracker could not be assigned to any relation (movement direction), or the relation is disabled.

• CrosssectionUncounted — The tracker did not cross any defined cross-section line.

• Cargo — The object was identified as cargo (e.g., load on a truck) rather than a countable vehicle.

• LabelUncounted — The tracker's label (category) is disabled from counting in the current configuration.

• Invalid — (Deprecated) Other reason for invalidation.

Classifications

GPR (Generalny Pomiar Ruchu) — General Traffic Measurement - the official Polish classification standard for traffic counting, defining vehicle categories for national road surveys.

Euro-6 — A European classification standard for vehicle categorization based on emission standards and vehicle types.

SCPR (Stacje Centralnego Pomiaru Ruchu) — Central Traffic Measurement Stations - a classification used by permanent traffic counting stations in Poland.

File Types

errata.pickle — A file storing manual corrections made using the Errata tool. Preserves original AI results while tracking user modifications.

config.toml — Configuration file for a measurement location, storing relations, cross-sections, classification settings, and other project parameters.