CodeProject.AI and Blue Iris: installation and alert setup

This tutorial, uploaded February 26, 2023, demonstrates CodeProject.AI 2.0.8 with Blue Iris on Windows. The speaker identifies that release as the latest at filming; no recording date or Blue Iris version is supplied. The settings reflect that period.

The walkthrough installs the AI server on the Blue Iris computer, connects it through the global AI settings, and enables detection for an individual camera. Blue Iris sends still images after motion triggers, and CodeProject.AI returns object detections. The speaker configures alerts to fire only after a confirmed match, aiming to reduce notifications caused by shadows or other irrelevant movement.

Camera tuning covers object labels, confidence thresholds and image timing. The demonstrated defaults are 50% confidence and two additional images at 750 ms intervals. The speaker suggests testing these first and explains how an early image can miss a vehicle still far down a driveway. More images increase processing load. Detection confidence is a model score, rather than a verified probability that an alert is correct. Detection confidence is a model score, rather than a verified probability that an alert is correct.

For this local AI setup, the speaker chooses CPU processing because his graphics card is underpowered and checks Windows Task Manager during detection. He also describes his separate Docker deployment on an Unraid server, recommends disabling automatic start/stop for a remote instance, and explains why static-object filtering sends periodic images even without motion. Custom models and license plate recognition receive brief introductions rather than full setup instructions.