Autonomous Drones Using VSLAM

Abstract

This thesis presents the development and evaluation of an autonomous drone system designed for indoor navigation and exit detection using Visual Simultaneous Localization and Mapping (VSLAM) techniques. The primary objective of this research is to enable drones to navigate and map indoor environments autonomously, leveraging advanced computer vision algorithms for real-time localization, mapping, and exit detection. The research integrates several key technologies: ORB-SLAM2 for visual odometry and mapping, hierarchical clustering for point cloud processing and exit detection, and custom path planning algorithms for autonomous navigation. The implementation utilizes a DJI Tello drone as the hardware platform, with software development primarily in Python, incorporating libraries such as djitellopy for drone control, Open3D for point cloud manipulation, and scikit-learn for clustering algorithms. Experimental results demonstrate the system’s efficacy in real-world indoor scenarios, highlighting its potential for applications in search and rescue operations, indoor mapping, and autonomous navigation in GPS-denied environments. The findings indicate significant improvements in navigation accuracy and environmental mapping capabilities compared to traditional GPS-based systems in indoor settings. However, the research also identifies several challenges, including limitations in battery life affecting mission duration, computational constraints necessitating off-board processing, and the need for robust performance under varying lighting conditions. These challenges provide direction for future work in the field. This research contributes to the growing body of knowledge in autonomous drone technology by providing insights into the practical integration of VSLAM, point cloud processing, and navigation algorithms for indoor environments. The developed system offers a foundation for further advancements in autonomous indoor drone navigation and exit detection. i