Quotations and deployment documents for AI cameras use many specialist terms: engine, object detection, ROI, dwell, anti-spoofing. This article explains each term the way Vaidio AI Vision uses it, so that everyone reading a quotation or talking to a vendor shares the same meaning. Infrastructure and licensing terms are in the companion article AI camera deployment terms.

AI foundations

Recognition AI (analytics AI, traditional AI)

AI that performs one defined task: recognizing objects, classifying, predicting. The output is a label or a number with a confidence level. Vaidio AI Vision belongs to this group; the detailed comparison with generative AI is in the article Generative vs traditional AI.

Generative AI

AI that creates new content (text, images) from learned data. Vaidio is not generative AI; LLM/VLM integration is an extension outside the Core platform, scoped per project.

LLM / VLM

Large language models and vision-language models. Both belong to the generative AI group; in the Vaidio system, LLM/VLM integration is an extension beyond the Core platform.

Engines and recognition

Engine

One AI analytics model enabled on a camera, for example face recognition, license plates or intrusion. Each engine is enabled per camera by license, and many engines run in parallel on the same video stream on Vaidio Core. See Multi-task AI camera.

Object Detection

The base layer shared by every engine: detecting and classifying people, vehicles, animals and objects in each frame, stored as metadata. Details on the Object recognition page.

Metadata

Structured data stored with each recognized object: type, color, count, time, camera. Metadata is what makes second-fast search across large video archives possible.

ROI (region of interest)

An area drawn on the camera frame to limit where recognition or alerts apply. Any shape, regions can overlap; noisy areas are marked as exclusion regions. Heavily used when configuring intrusion alerts.

Human Attributes

The person-attributes engine: clothing, age group, gender. A separate engine with its own license; aggregated age and gender figures are viewed in Vaidio Data.

LPR / ANPR (license plate recognition)

Two general names (License Plate Recognition / Automatic Number Plate Recognition) for the same technology that reads plates. This website uses the name License plate recognition.

Behavior and events

Dwell (dwell time)

How long an object stays inside a monitored region. Setting a dwell threshold powers loitering alerts (1–1,200 seconds) or waiting-time measurement; used with unusual behavior detection and heatmaps.

Occupancy

How many people or vehicles are present in an area at a given moment, updated in real time. Unlike pass-through counting, occupancy answers “how many are here now”, used mainly in crowd coordination.

Object Counting

Counting passes across a counting line by direction in and out, up to 10 object types per line; counting accuracy of 95%. See Traffic monitoring.

Object Left Behind

A dedicated engine that raises an alert when an object sits in a region without a person for longer than the configured time (1–3,600 seconds); object types are selected in advance. Different from scene change detection. See Left-behind object detection.

Scene Change Detection

Compares the current frame against a background model of the normal scene, without declaring object types in advance. Outdoors it reports fallen trees, falling rocks, landslides and flooding. See Scene change detection.

Hashtag

A tag attached to an event so the exact situation can be recalled with one keyword, for example #accident. Tags are available on every search result and group same-type events during investigations.

Identity and privacy

Anti-Spoofing

The engine that checks whether a face in front of the camera is a live person or a printed photo, or a picture or video played on a screen, labeling each recognition attempt as real or fake. Requires its own license alongside the face recognition license. See Face recognition.

PPE (personal protective equipment)

The engine that checks helmets and reflective vests (present/absent) for each person in the frame; face masks are checked through the face recognition engine. See PPE recognition.

Privacy Protection

An optional toggle: when enabled, faces in search results, alerts and exported images are blurred; only accounts with permission can view original images. Related to personal data blurring.

Natural Language Enhancement (NLE)

The engine that matches imagery against natural language descriptions for search and alerts, such as “a person carrying a bottle”. Requires its own license; belongs to recognition AI, not generative AI. Explained fully in Generative vs traditional AI.

Traffic Management

The module that tracks speed and tags speeding and red-light violations; enabled with its own license, and cameras need a clear view of the traffic signals. Industry use case in Smart traffic.

Continue with AI camera deployment terms: on-premise, edge, VMS, ONVIF, license, RMA/DOA and other system words. Or go back to AI Vision resources to pick another article.