Monocular Snapshot-based Sensing and Control of Hover, Takeoff, and Landing for a Low-cost Quadrotor

What is it about?

This paper mainly concerns the hover control of a quadrotor. A “snapshot” paradigm is utilized that takes and stores an image of the ground under the MAV as a reference image, and the position of the vehicle can then be estimated from the transformation of the following images with respect to the reference image so that long-term drift can be contained. I have implemented and tested this approach on the Parrot AR Drone 1.0. In my algorithm, real-world challenges have been considered, namely the illumination change and external disturbance. Based on an open-source program, all the image processing and control algorithms run on-board the vehicle. In the project, a fast and robust template matching algorithm is developed. I managed to estimate the attitude, speed and position from the raw and noisy IMU and image data using a simple complementary filter.

Featured Image