AI Glossary

The simulation of human intelligence by computer systems, including learning from data, reasoning about problems, and making decisions autonomously.

AI systems that operate with a degree of autonomy, planning multi step tasks, making decisions, and executing workflows without constant human supervision.

An autonomous software entity that perceives its environment, makes decisions, and takes actions to achieve specific goals using AI capabilities.

Automatic Number Plate Recognition (also called Automatic License Plate Recognition) uses AI to read vehicle registration plates from camera feeds in real time.

An AI assistant integrated into a specific workflow or application that provides contextual suggestions, automates routine tasks, and augments human decision making.

A field of AI that enables computers to interpret and understand visual information from images and video, extracting meaningful data from pixel input.

AI analysis that estimates the number of people and density levels in a given area, used for public safety, venue capacity management, and stampede prevention.

The maximum amount of text (measured in tokens) that an AI model can process in a single interaction, determining how much information it can consider when generating responses.

A subset of machine learning that uses neural networks with many layers to learn complex patterns from large amounts of data, particularly effective for image and video analysis.

Numerical representations of data (text, images, audio) in high dimensional vector space where similar items are positioned closer together, enabling semantic search and comparison.

Running AI inference directly on local devices near the data source rather than in a centralized cloud, reducing latency, bandwidth costs, and privacy concerns.

Processing data near where it is generated rather than sending it to a centralized data center, enabling faster response times and reduced bandwidth usage.

A large AI model trained on broad data that can be adapted to a wide range of downstream tasks, such as GPT, Claude, Gemini, and LLaMA.

An AI capability that identifies or verifies individuals by analyzing facial features from images or video. Visylix achieves 99.7% accuracy with liveness detection.

The ability to search recorded video across multiple cameras using AI criteria such as face, vehicle plate, time range, object type, or event type to quickly find relevant footage.

The process of further training a pre trained AI model on domain specific data to improve its performance on particular tasks or knowledge areas.

AI systems that can create new content including text, images, code, music, and video based on patterns learned from training data.

Using Graphics Processing Units to accelerate AI model inference and training, providing significantly faster processing compared to CPUs for parallel computing workloads.

A visual representation of data using color gradients to show areas of high and low activity, commonly used in retail to understand customer movement and dwell time patterns.

HTTP Live Streaming is an adaptive bitrate streaming protocol developed by Apple that delivers video over standard HTTP connections, widely supported across devices and browsers.

When an AI model generates information that sounds plausible but is factually incorrect or fabricated, a common challenge with language models that RAG techniques help mitigate.

The process of running a trained AI model on new data to generate predictions or outputs, as opposed to training where the model learns from data.

A type of foundation model trained on massive text datasets that can understand and generate human language, power chatbots, and assist with complex reasoning tasks.

Low Rank Adaptation is a parameter efficient fine tuning technique that adds small trainable matrices to a frozen model, dramatically reducing the compute required for customization.

A subset of AI where systems learn patterns from data and improve performance over time without being explicitly programmed for each task.

AI models that can process and understand multiple types of input simultaneously, such as text, images, audio, and video, reasoning across modalities.

A standardized protocol for connecting AI models to external tools, databases, and services, enabling AI agents to interact with real world systems.

AI powered analysis that identifies movement within video frames using adaptive background modeling, shadow suppression, and intelligent filtering to reduce false alarms.

A technique that reduces the precision of model weights (e.g., from 32 bit to 8 bit) to decrease model size and improve inference speed while maintaining most of the accuracy.

A compression technique that removes unnecessary connections and parameters from a neural network to create a smaller, faster model with minimal accuracy loss.

Machine Learning Operations combines ML, DevOps, and data engineering practices to deploy, monitor, and maintain ML models in production reliably and efficiently.

A computing system inspired by the biological neural networks of the human brain, consisting of interconnected nodes that process information in layers.

A hardware or software system that records and stores video from IP cameras over a network, typically providing basic playback and management capabilities.

A family of compact, power efficient AI computing modules designed for edge deployments, capable of running multiple AI models simultaneously for robotics, video analytics, and IoT.

An AI technique that identifies and locates specific objects within images or video frames, drawing bounding boxes around detected items with classification labels.

Open Network Video Interface Forum is an industry standard protocol that ensures interoperability between IP cameras and video management systems from different manufacturers.

An Intel toolkit for optimizing and deploying AI inference on Intel hardware including CPUs, integrated GPUs, and VPUs, enabling efficient AI processing without discrete GPUs.

AI technology that follows individuals across multiple camera views, maintaining identity through re identification even when subjects temporarily leave the frame.

AI technology that detects human body positions by identifying key skeletal points (typically 17 keypoints), used for fall detection, ergonomic analysis, and behavior recognition.

Personal Protective Equipment detection uses AI to verify whether workers are wearing required safety gear such as hard hats, safety vests, goggles, and gloves.

The practice of crafting effective instructions and context for AI models to produce desired outputs, a critical skill for building applications on top of foundation models.

A technique that enhances AI responses by retrieving relevant information from external knowledge bases before generating an answer, reducing hallucinations and improving accuracy.

The ability to recognize and match the same person across different camera views and time periods without relying on face recognition, using body shape, clothing, and gait analysis.

Real Time Streaming Protocol is a network protocol used to control streaming media servers, commonly used to deliver video from IP cameras to recording and management systems.

Real Time Messaging Protocol is a streaming protocol originally developed by Adobe, widely used for live video broadcasting and ingestion.

Reinforcement Learning from Human Feedback is a training technique where AI models are refined using human preferences to improve helpfulness, accuracy, and safety.

A compact language model optimized for efficiency that runs on edge devices or limited hardware while retaining strong performance for specific domain tasks.

Secure Reliable Transport is an open source streaming protocol that delivers high quality, low latency video over unreliable networks, handling packet loss and jitter.

The capability of AI models to invoke external tools such as APIs, databases, and calculators to gather information and take actions beyond their training data.

An NVIDIA SDK for high performance deep learning inference that optimizes AI models for deployment on NVIDIA GPUs, delivering lower latency and higher throughput.

The basic unit of text that AI language models process, typically representing a word, part of a word, or punctuation mark. Model pricing and context windows are measured in tokens.

An AI capability that counts distinct individuals rather than total entries, using re identification to avoid counting the same person multiple times.

A specialized database designed to store and efficiently search high dimensional vector embeddings, commonly used in RAG pipelines and semantic search applications.

Software platform that manages video surveillance cameras, handles recording, provides live monitoring, and enables playback and forensic search across camera networks.

Web Real Time Communication is a technology that enables ultra low latency peer to peer video streaming directly in web browsers without plugins. Visylix achieves sub 500 millisecond latency.

A system design that minimizes data copying operations during video processing, dramatically improving throughput and reducing memory usage for high scale deployments.