YOLOv8n-Driven Cursor Control

I need a compact C++ application that continuously captures my primary monitor, runs each frame through the YOLOv8n model, and locates any human-shaped objects that appear. As soon as a target is detected, the program should calculate the centre of that bounding box and stream simple mouse-movement commands over a USB serial link to an attached Arduino. The Arduino will then translate those commands into actual cursor motions on the same PC.

I already have a YOLOv8n model fine-tuned for these human silhouettes, so no training pipeline is required—just load the supplied weights. Your job is to:

• build the screen-grab + inference loop in modern C++ (OpenCV and the official Ultralytics YOLOv8 C++ API or equivalent ONNX/TensorRT implementation are fine)
• design a minimal, reliable serial protocol (9600 bps or faster) that carries x/y delta values and optional click flags to the Arduino over USB
• provide both the desktop source/executable and a short, matching Arduino sketch that receives the data and drives the Mouse library

Please structure the code so model-loading paths and serial port settings can be tweaked easily, and include a brief README describing build steps, required libraries, and how to run everything end-to-end. As long as detections fire correctly and the cursor follows the bounding-box centre without noticeable lag, we’re good.