Key terms and definitions used across the AgenticEye platform and the broader AI vision industry.
A category of artificial intelligence systems that can autonomously perceive, reason, and take actions to achieve defined goals. In the context of AgenticEye, agentic AI refers to intelligent agents that continuously monitor camera feeds, detect events, and trigger automated responses without requiring constant human supervision.
The process of identifying patterns, events, or observations in video data that deviate significantly from expected behavior. Anomaly detection enables security systems to flag unusual activity such as unauthorized access, abandoned objects, or atypical movement patterns.
A rectangular border drawn around a detected object in an image or video frame. Bounding boxes are used to indicate the location and approximate size of identified objects, people, or vehicles within a scene, and are a fundamental output of object detection models.
A continuous stream of video data transmitted from a surveillance camera to a monitoring system. Camera feeds can be delivered over IP networks using protocols such as RTSP or ONVIF and serve as the primary input for AI-powered video analytics.
A field of artificial intelligence that enables machines to interpret and understand visual information from images and video. Computer vision encompasses tasks such as object detection, image classification, facial recognition, and scene understanding.
A class of deep neural network architectures specifically designed for processing visual data. CNNs use convolutional layers to automatically learn spatial hierarchies of features from images, making them highly effective for tasks like image classification, object detection, and segmentation.
A distributed computing paradigm where data processing occurs close to the source of the data, such as directly on a camera or a local gateway device, rather than in a centralized cloud data center. Edge computing reduces latency, conserves bandwidth, and enhances privacy for real-time video analytics.
An alert or detection triggered by an AI system when no actual event of interest has occurred. For example, a security system might generate a false positive by misidentifying a shadow or an animal as a person. Minimizing false positives is critical for maintaining operator trust and reducing alert fatigue.
The number of individual video frames captured or processed per second. Higher frame rates provide smoother video and more temporal detail for AI analysis, but require greater computational resources. Common surveillance frame rates range from 15 to 30 FPS.
A technique that uses virtual geographic boundaries to define zones within a camera's field of view or across a physical space. When an object or person enters, exits, or dwells within a geofenced area, the system can trigger alerts or automated actions.
The use of graphics processing units (GPUs) to speed up the computation required for AI inference and model training. GPUs excel at the parallel mathematical operations central to neural network processing, enabling real-time analysis of multiple video streams simultaneously.
The task of assigning a label or category to an entire image based on its visual content. Unlike object detection, which locates and identifies multiple objects within a scene, image classification determines what a single image predominantly depicts, such as "parking lot," "warehouse," or "retail floor."
The process of running a trained machine learning model on new, unseen data to generate predictions or detections. In video analytics, inference occurs when a deployed model analyzes live camera frames to detect objects, classify scenes, or identify anomalies in real time.
The time delay between when a video frame is captured and when the AI system produces a result or alert for that frame. Low latency is essential for real-time monitoring applications where rapid response to security events is critical.
A subset of artificial intelligence in which algorithms learn patterns and make decisions from data without being explicitly programmed for every scenario. Machine learning models improve their accuracy over time as they are exposed to more training data and feedback.
The process of teaching a machine learning model to recognize patterns by exposing it to large datasets of labeled examples. During training, the model adjusts its internal parameters to minimize prediction errors, ultimately learning to generalize from training data to new, real-world inputs.
A computing architecture inspired by the structure of the human brain, consisting of interconnected layers of nodes (neurons) that process and transform data. Neural networks form the foundation of modern deep learning and are used extensively in computer vision for feature extraction and pattern recognition.
The computer vision task of identifying and locating specific objects within an image or video frame. Object detection models output both the class (e.g., person, vehicle, package) and the spatial location (typically as a bounding box) of each detected object.
Open Network Video Interface Forum — an industry standard protocol that enables interoperability between IP-based security cameras and video management systems from different manufacturers. ONVIF compliance allows AgenticEye to integrate with a wide range of camera hardware.
A technology that extracts text from images or video frames. In surveillance and monitoring contexts, OCR is commonly used to read license plates, signage, labels, and other printed or displayed text captured by cameras.
The task of recognizing and tracking the same individual across different camera views or after temporary occlusion. Person re-identification uses appearance-based features such as clothing color, body shape, and gait to match a person seen in one camera to their appearance in another, without relying on facial recognition.
The ability to analyze and respond to video data as it is captured, with minimal delay. Real-time processing ensures that alerts and detections are delivered within seconds of an event occurring, enabling immediate human response or automated actions.
A user-defined area within a camera's field of view where the AI system focuses its analysis. By limiting detection to specific regions, ROIs reduce false positives from irrelevant areas and improve processing efficiency by ignoring portions of the frame that do not require monitoring.
Real-Time Streaming Protocol — a network protocol used to establish and control the delivery of video streams from IP cameras. RTSP enables AgenticEye to connect to and receive live video feeds from compatible cameras for real-time AI analysis.
A computer vision technique that classifies every pixel in an image into a predefined category, producing a detailed map of the scene. Unlike object detection, which draws bounding boxes, semantic segmentation provides precise outlines of objects, surfaces, and regions, useful for understanding complex environments.
A multi-dimensional array of numerical data used as the fundamental data structure in deep learning frameworks. Images are represented as tensors (with dimensions for height, width, and color channels) before being processed by neural networks, and all intermediate computations flow through tensor operations.
The automated analysis of video content using AI and computer vision algorithms to extract meaningful information, detect events, and generate actionable insights. Video analytics transforms passive camera footage into an active intelligence layer for security, operations, and business optimization.
You Only Look Once — a family of real-time object detection models known for their speed and efficiency. YOLO architectures process an entire image in a single pass through the neural network, enabling high-speed detection suitable for live video analytics and edge deployment scenarios.