In the context of the project, a Python application was implemented which can receive images from the drone camera in real time, and allow drone control through a computer keyboard (take off, land, move forward, turn left/right). All received information, together with the commands to the drone are saved in a file. Using these features, a Convolutional Neural Network is trained over the dataset created from the recorded images and actions. Th final goal is to be able to provide an autonomous visual navigation system for drones that will follow a line drawn to the ground.

Using the Python application, the drone flew over the black line created on the floor, using the keyboard navigation, while recording the series of control commands and images. The was repeated several times, and the dataset created was used to repetitively train the model. After several iterations, the result was satisfactory.