Hikvision Refines Video Surveillance with Large AI Models

A year after debuting its Guanlan large-scale AI models, Hikvision is shifting video surveillance from reactive monitoring to predictive intelligence. By integrating multimodal processing directly into cameras and management platforms, the company claims to have slashed false alarms by up to 90% while automating complex administrative and search tasks.

Corp and Tech·September 8, 2026·reads: 796

Hikvision Refines Video Surveillance with Large AI Models

The Guanlan architecture addresses the persistent issue of false alerts in perimeter security, where environmental factors like weather or wildlife often overwhelm legacy systems. By deploying DeepinViewX cameras that process data at the edge, Hikvision reports a 90% reduction in false alarms and double the detection range of traditional hardware. These models extend beyond simple object detection into scenario-specific analytics, including PPE compliance on construction sites, fall detection in healthcare facilities, and forklift monitoring in logistics hubs.

Efficiency gains extend to data management and user interaction. The Guanlan Encoding system uses region-of-interest segmentation to prioritize dynamic subjects, such as people or vehicles, over static backgrounds, yielding storage savings of 30% to 50%. Meanwhile, the AcuSeek tool enables operators to query video archives using natural language, replacing manual timeline scrubbing with descriptive searches.

Interaction is further evolving with the introduction of the HIKO AI agent, which integrates with the HikCentral Professional platform. This interface functions as an administrative assistant, a security guardian, and a search engine. Users can now generate attendance reports, configure perimeter rules, or reconstruct a person's movement path across multiple data points using simple conversational commands rather than complex manual configurations.