With the advancement of technology, human-robot interaction (HRI) is becoming more intuitive, including through hand gesture-based control. This study aims to develop a real-time hand gesture recognition system to control a quadcopter swarm within a simulated environment using ROS and Gazebo. The system utilizes Google's MediaPipe framework for detecting 21 hand landmarks, which are then processed through a custom-trained neural network to classify 13 predefined gestures. Each gesture corresponds to a specific command such as basic motion, rotation, or swarm formation, and is published to the /cmd_vel topic using the ROS communication framework. Simulation tests were performed in Gazebo and covered both individual drone maneuvers and simple swarm formations. The results demonstrated a gesture classification accuracy of 90%, low latency, and stable response across multiple drones. This approach offers a scalable and efficient solution for real-time swarm control based on hand gestures, contributing to future applications in human-drone interaction systems.