The increasing demand for efficient and autonomous healthcare logistics in remote and inaccessible regions has driven advancements in vision-based navigation for Micro Aerial Vehicles (MAVs). This research presents a fully vision-based object detection system for MAVs, eliminating the need for additional sensors such as LiDAR or GPS. The proposed system leverages the YOLOv8 model to enable real-time detection, target identification, and obstacle avoidance. A structured methodology, including dataset preparation, annotation, model training, and evaluation, ensures high accuracy and robust performance. The system achieves a mean Average Precision of 94.9% in multi-class detection and operates effectively in real-time environments. Experimental results demonstrate the feasibility of deploying a vision-only navigation framework for autonomous medicine delivery, reducing hardware complexity and operational costs. The proposed approach enhances scalability, making it suitable for broader applications in disaster relief, surveillance, and smart logistics.
Keywords— Vision-based navigation, object detection, MAV, YOLOv8, autonomous medicine delivery.