Enhancing path following drone : using image-based sensor matrix

Sammanfattning: Drones are a fascinating form of transport and have the potential to dominate transportation in the future. Drones are usually faster and are unstable when flying in the air. Drones can be automated in many ways to realize various applications, one of which is a path following drone. Path following drones can follow a path that can be distinguished from the background. This kind of drone finds many applications for example following rivers or roadways for mapping, surveillance in the military, and medicine or blood transportation in healthcare. Our thesis aims to implement a new technique and evaluate the performance of the path following drone. The main objective is to make the drone more responsive to sudden changes in the path. To achieve this we employed a higher-order matrix analysis of the image retrieved from the drone. Throughout the thesis the image from the drone is divided into a 3x3 matrix of sections and each section is assigned a value. These matrices are further analyzed to plan the movement of the drone.  The concept of masks was deployed which reduced the computation to a great extent as compared to look-up tables with all possible matrices. Each mask is again a 3x3 matrix and represents a particular direction and speed of the drone. Each time all masks are applied to the image from the drone and the mask which is closest to the drone image matrix is chosen and the drone is controlled accordingly to the closest mask. We also find the contour of the path regularly and find the center of such contour. This center helps the drone to get back to the course of the path if the drone is going too far away from the path. The algorithm is coded in Python programming language and the OpenCV library is used for image processing tasks.  The results of our work show that we could improve the performance of the path following drone. Though the performance of the drone is limited by some factors such as the drone’s bias in a particular direction, lighting conditions, etc. Also, the drone’s performance is inhibited by the choice of parameters like threshold, speed, and angle vectors. Overall the drone is found to be responsive to rapid changes in the path with the implementation of higher-order matrix analysis, contouring, and the concept of masks.