Controle de posicionamento de robô aéreo autônomo utilizando visão computacional

Abstract

The use of autonomous aerial robots, popularly known as drones, has been growing and revolutionizing the way humans interact with airspace. In recent years, these robots are gaining in notoriety and possibilities of use, due to the improvement of their performance capabilities, as a result of the improvement of microelectronics, which have managed to miniaturize electronic components, sensors and cameras. The growing access to drones, combined with academic research, has made possible a scenario of commercial applications, such as mapping, environmental monitoring, package delivery, filming, among many others. For these drones to navigate in unstructured environments, avoiding collision with obstacles and in a stable and autonomous way, several techniques and technologies are necessary, such as Computer Vision, feedback control systems and embedded systems, which will be addressed in this work. Considering this scenario, this work proposes and develops an autonomous tracking and following system for a drone in relation to a target, which consists of processing the images from the drone's on-board camera, identifying the face of a person to be tracked, and maintain a predetermined distance in relation to the person, who can move freely in the environment. Images are sent from the drone (DJI Tello model) to a computer via WiFi communication. These images are processed with the MediaPipe library (used for image segmentation and classification) which determines the target to be tracked and followed. The automatic position control system of the drone is implemented with a Proportional-Integrative-Derivative (PID) controller. The work presents the detailed configuration and assembly procedure of the project, in addition to the practical results obtained with the execution of the proposed task.