Hikvision advances Guanlan AI models for video surveillance

Hikvision has reported advancements in its Guanlan large-scale AI models, one year after their initial launch. The technology is designed to enhance video surveillance systems by providing deeper situational understanding through natural language innovations and improved detection accuracy.

Key applications include perimeter protection, where DeepinViewX cameras utilizing the Guanlan model can reduce false alarms by 90% and detect targets at twice the range of traditional cameras. The system also offers scenario-specific analytics for safety and compliance, such as detecting Personal Protective Equipment (PPE), monitoring for falls in care homes, and identifying smoking in gas stations. Other operational uses include forklift detection in logistics parks and fire or smoke detection in mining sites.

To optimize efficiency, the Guanlan Encoding engine uses Region of Interest (ROI) segmentation. This process preserves high detail on dynamic subjects like people and vehicles while applying heavier compression to static backgrounds, thereby reducing storage requirements.