This paper presents a novel framework for the autonomous landing of quadcopters using a vision-based visual servoing controller. The proposed system integrates advanced computer vision algorithms and model predictive control (MPC) to enable precise, smooth, and energy-efficient landings on platforms, even under external disturbances and complex backgrounds. AprilTag-based visual markers are detected using Region-based Convolutional Neural Networks (RCNN), while HOG and SIFT features are used to estimate position and orientation with sub-decimeter accuracy. The extracted visual information feeds into the MPC module, which generates an optimized trajectory considering control input constraints, energy consumption, and obstacle avoidance. Simulation results reveal that the proposed MPC controller significantly outperforms PID and sliding mode control (SMC) methods in terms of path tracking accuracy, energy consumption, and disturbance rejection. The system maintains position errors under 5 cm and reduces cumulative energy use by more than 50% compared to conventional methods. Moreover, real-time processing and visual feedback ensure robust performance in environments with grass, lighting changes, and wind gusts. Sensitivity analysis confirms that the controller remains stable under parameter variations. These findings demonstrate the effectiveness of combining deep visual recognition with predictive control for high-precision autonomous UAV landing missions.

Key words: Model Predictive Control, Visual Servoing, Quadcopter, Autonomous Landing, Computer Vision, Deep Learning, AprilTag.