Video analytics is software that examines video from security cameras to detect, classify and track things in the scene, such as people, vehicles or objects, and to flag events like someone crossing a line or lingering in a restricted area. It also indexes recorded video so staff can search for, say, a person in a red jacket or a white van instead of scrubbing through hours of footage. Many current systems use machine learning, which is why it is often marketed as AI video analytics.
At a glance
- Analytics turn raw video into detections, alerts and searchable records of people, vehicles and events.
- Common features include object classification, line crossing, loitering, intrusion zones, crowd or occupancy counts and license plate reading.
- Processing may run on the camera, a recorder or server, or in the cloud, depending on the product.
- Accuracy depends heavily on camera placement, lighting and scene; test before relying on it.
- Facial recognition and other biometric uses carry legal requirements that vary by jurisdiction.
What problem it solves
Most camera systems record far more video than anyone can watch. Basic motion detection triggers on shadows, rain, headlights and swaying trees, so alerts get ignored. After an incident, staff may spend hours reviewing footage across several cameras to find a few seconds that matter.
Video analytics narrows that down. By recognizing that an object is a person or a vehicle, it can cut nuisance alerts, notify staff or a monitoring center when someone enters a restricted area after hours, and let investigators search across cameras by object type, color, time or location. Some organizations also use it for operational purposes such as counting visitors, measuring queue length or occupancy monitoring.
How it works
Detection and classification. A model analyzes each frame or a sample of frames, finds objects and labels them, typically as person, vehicle or other categories the product supports. It then tracks objects between frames to understand movement.
Rules and alerts. Administrators draw zones, lines or schedules on a camera view. The system generates an event when a tracked object meets a rule, such as crossing a virtual line in a set direction, staying in an area too long (loitering) or entering a zone after hours.
Search and metadata. Many systems store descriptive metadata (object type, colors, direction, time) alongside video so recorded footage can be searched. Some newer products support natural-language search, where you type a description and the system returns matching clips; results vary by product.
Specialized analytics. License plate recognition (LPR) reads plate characters; other modules count people, detect smoke or fire, or flag objects left behind. Some products offer facial recognition, which is legally sensitive.
Where it runs. Analytics can run on IP cameras with built-in processors, on a network video recorder (NVR) or analytics server, or in a cloud platform such as video surveillance as a service (VSaaS). Running at the edge reduces bandwidth; cloud processing can make updates and cross-site search easier.
When it matters for buyers
- When false alarms are drowning real ones. Object-based detection is often the first upgrade over motion alerts.
- When video is monitored remotely. Monitoring centers commonly rely on analytics to decide which events need a person to look.
- When investigations take too long. Forensic search across cameras can save significant staff time, depending on how well it works on your footage.
- When choosing cameras. Analytics features often depend on the camera model and vendor, so they should be part of camera selection.
- When privacy or biometrics is involved. Video of identifiable people can be personally identifiable information (PII), and face or other biometric processing is regulated in some states and countries.
Questions to ask vendors
- Which analytics run on the camera, which on a server or recorder, and which in the cloud?
- What are the licensing costs per camera or per feature?
- Can we trial the analytics on our own cameras and sites before buying?
- How do you measure and report false alarms and missed detections?
- Do any features identify individuals, such as facial recognition, and can they be disabled?
- Where is video and metadata processed and stored, and for how long?
- How do analytics events integrate with our alarm monitoring, access control or incident tools?
How it differs from basic motion detection
Basic motion detection flags changes in pixels, so it reacts to anything that moves or changes brightness, including weather, lights and animals. Video analytics tries to understand what is moving, such as a person or a vehicle, and applies rules to that object’s behavior. That usually means fewer nuisance alerts and better search, though results still depend on the product, camera and scene. For help comparing camera and analytics options, see our video surveillance overview.