SiteSignal — Visual Safety at the Edge

SiteSignal is a privacy-first visual safety system designed for warehouses, factories, construction sites, laboratories, and other workplaces with controlled or hazardous areas. The deployed web application performs real person detection locally in the browser using TensorFlow.js and COCO-SSD. Operators can define warning and restricted safety zones, while SiteSignal’s geometry engine continuously evaluates detected person bounding boxes against those zones. The system automatically changes between SAFE, WARNING, and BREACH states and records incidents with timestamps and detection information. It also provides an incident timeline, analytics, local persistence, and data export. SiteSignal does not use facial recognition and does not upload raw workplace video to a cloud inference service by default. For the Qualcomm Model-to-Device Innovation track, we also successfully compiled, profiled, and tested YOLOX Small through Qualcomm AI Hub on a hosted Snapdragon X Elite CRD running Windows 11. Qualcomm AI Hub reported 14.4 ms estimated inference time, 14 MB estimated peak memory, and all 296 model operations executing on the NPU. SiteSignal is designed as a safety assistance layer rather than a replacement for trained safety professionals or certified safety systems.