Event-based weapon detection for a jeweller

01Situation

The store already had cameras. They recorded footage but only helped after the fact. In a robbery, seconds count, and staff have no free hand to raise the alarm.

02Challenge

Alert faster without monitoring customers. Person-based video analytics was ruled out for privacy reasons and because of the customer experience. No new camera hardware was to be required.

03Solution

The AI analyses footage from the existing cameras and looks only for weapons: objects, not people. When a weapon is detected, predefined steps are triggered: a silent alarm, time-limited recording and a set escalation path. A person reviews every alert.

04Implementation

• —Audit of existing cameras: positions, image quality, views of entrance and sales floor
• —Define zones, alert paths and responsibilities together with the client
• —Connect detection to the existing camera streams, with no new cameras
• —Silent alarm so staff don't have to act themselves in an emergency
• —Data minimisation: no biometric data, event-based storage only, with retention limits
• —Test phase to tune false alarms and procedures before going live

05Impact

• —Alerts raised without staff having to act
• —Faster response in an emergency instead of footage for later
• —No impact on the customer experience: no facial recognition, no profiling
• —Complements existing security measures; no system can guarantee safety