Development of vision-based drones for obstacle avoidance

• Main Author: Lee, Josiah Rong Guang
• Other Authors: Mir Feroskhan
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Mechanical engineering
• Online Access: https://hdl.handle.net/10356/166795
• Institution: Nanyang Technological University

Drones, commonly known as unmanned aerial vehicles (UAVs), are a type of aircraft which is operated without the assistance of any human pilot on board. Drones have revolutionized a wide range of industries, from agriculture and logistics to environmental monitoring and emergency response. The safe and autonomous navigation of drones is paramount to maximize their potential and reduce risks associated with collisions and accidents. This project will focus on the development of a vision-based drone for obstacle avoidance, thereby, addressing this crucial aspect of drone operation. The proposed system in this project harnesses the power of computer vision and deep learning techniques to enable drones to perceive their environment and adapt their flight trajectory accordingly. The author has made use of computer vision to detect and classify various types of obstacles from camera feeds in real-time and the algorithm is further enhanced by incorporating motion estimation and object tracking to avoid potential threats in dynamic environments. This vision-based drone system significantly improves the overall safety, reliability, and autonomy of drone operations, allowing for seamless integration into an ever-growing range of applications. This advancement is particularly important for operations in complex and cluttered environments, such as urban settings and disaster-stricken areas, where conventional ground position systems (GPS) and sensor-based navigation systems may not be sufficient or reliable. The algorithm used in this project has displayed positive results in various simulation environments, thereby showcasing its effectiveness and robustness.