What is flyguard?

Flyguard is a Python project for an autonomous drone system designed to kill horn flies on cattle farms. Horn flies are small parasitic insects that feed on cattle and cause weight loss, reduced milk production, and disease. The system flies a drone on a patrol route over a pasture, uses a camera and AI object detection to spot horn flies on cattle, and fires a laser aimed by a motorized mount to kill them. The safety design is the most detailed part of the codebase. The laser can only fire when every one of a set of independent checks passes: the system must be armed with a hardware key, the drone must be within a defined geofence, no human must be visible in the camera frame, the detected target must be identified as a horn fly with sufficient confidence, the beam must be aimed clear of the cattle's head and eyes, the distance must be within a safe range, the gimbal must have settled on the target, a thermal duty cycle limit must not be exceeded, and a live heartbeat signal must confirm the operator is still present. Any failure in any check blocks firing. Every shot and every refusal is written to a tamper-evident audit log. The README notes that the laser involved is an eye hazard classified as Class 3B or 4. The system targets DJI Enterprise drones such as the Matrice 350 RTK as the primary hardware, paired with an NVIDIA Jetson Orin computer mounted on the drone. A fully simulated mode is included so the complete detection-fire-patrol loop can be tested without any physical hardware. A web-based ground control station runs locally and shows the live camera feed with detection overlays, a safety interlock panel that must show all green before firing is permitted, and controls for arming the patrol, disarming, and triggering an emergency stop. The emergency stop is designed to be reachable in one click from any state. The AI detection component uses YOLO, a common object detection framework, and the repository includes tools to generate synthetic training images and train the detector from scratch if you do not have labeled real-world data. The project is MIT licensed for the software, but the README notes that operating lasers and drones carries separate legal obligations that the software license does not cover.