mlweber/taubenturret-backend

taubenturret-backend is the server-side software for an automated pigeon deterrent system called TaubenTurret, whose name comes from the German word for pigeons. A camera spots pigeons, this backend identifies them using computer vision, and the physical turret device then activates to scare them away. The backend runs a REST API, a standard way for software components to talk to each other over a network, that accepts incoming images and returns object detection results. It uses YOLO, a widely used computer vision model for spotting and labeling objects in images, paired with OpenVINO, an Intel toolkit that converts those models into a highly optimized format that runs quickly on ordinary CPUs without needing a dedicated graphics card. It uses INT8 precision, a technique that reduces the size and computational cost of the model with minimal accuracy loss, so detection stays fast even on modest hardware. A built-in feature automatically saves every image that comes in for detection, so you can manually review later which detections were correct and which were mistakes. Over time this collection of labeled images lets you train a custom model fine-tuned to your specific camera setup and environment. The project is packaged as a Docker container, so it can run on most machines without complicated setup. Setting it up locally without Docker requires Python 3.10 or newer, plus make and uv for environment management, and it needs the YOLO weights exported into the OpenVINO format before first use, which by default downloads the full COCO dataset. Once running, it exposes four API endpoints, a health check, a list of supported object classes, and detection across all classes or a specific subset such as birds or cats, and includes a built-in Swagger UI for testing them. You would use this if you are building or running a TaubenTurret system and need the software side to process camera images and decide whether a pigeon is present. The project is written in Python.