- Video analytics on existing cameras: the chain from RTSP stream to event Computer vision How to set up real-time analytics without replacing the IP cameras already in the field: stream intake, GPU frame decoding, detection and tracking, rule engine and event output. A five-link chain that has settled over five years of deployments.
- Getting people counting to 96%: field calibration notes Computer vision A model that scores 99% in the lab drops to 85% in the field. What closes the gap is not the model but the camera angle, the position of the counting line, night data and the measurement method. Notes kept on the way to 96.2% in stores.
- Combining IoT and camera analytics: MQTT and an event-driven architecture IoT A camera is a sensor, the richest one. When it joins the same event bus as temperature, door contacts and PLC signals, information appears that none of them could produce alone. The event schema, MQTT topic structure and integration lessons we use in the field.
- Keeping detection models alive in the field: drift, retraining and dataset discipline Artificial intelligence Training the model is the shortest part of the project. The hard part is keeping a model that runs in 40 locations accurate for months while season, light, uniforms and layout change. The data and versioning discipline we settled on over five years.
Older articles (2018–2021) on HTML5, PowerShell, VBScript, SQL Server and Active Directory are available in Turkish only. Türkçe →