Manoeuvring drone (Tello and Tello EDU) using body poses or gestures

The project aims to use computer vision and machine learning to control Tello and Tello Edu Drones. The main use of this project is for educational and experimental purposes for upcoming future uses. I got inspired by the developments in the field of computer vision and pose estimation. As people are doing some much instructing and innovative things using this Drone and combined it with AI, Computer Vision, and Machine learning. This project helps anyone who likes to learn from the beginning even they don't have any prior knowledge of programming or Computer Vision. They can able to build a simple application and also provide the basic knowledge for the development of a more complex application. As a final objective, the main application is proposed, are consists of a drone that can be controlled using body poses or the movement of the body i.e., following the person by maintaining a constant safety distance and following the commands given by the instructor. To start programming Tello or Tello EDU drone, it is based on a series of scripts and functions which is already been developed. So, that SDK is used as a guide to developing a new tello custom SDK to develop a complex and interactive application. By using this custom template, we can develop several basic mission scripts/programs. For the main application, different experiments were carried out to check which method is better in extracting the body key points/Landmarks in real-time in which low computing power is required or in other words no or less GPU power is required. As a result, I come across the Google AI open source "MediaPipe Machine Learning" platform, which provides a cross-platform, customizable ML solution for live and streaming media. Which is used for advanced application in this project.

https://upcommons.upc.edu/handle/2117...