As a Thesis design, I have developed an Android application, which is capable to detect the cones used to mark the track at a Formula Student Driverless competition and furthermore is capable to build a map from the locations of these cones. In this thesis I present the most common approaches to the vehicle control in the field of mobile robots which includes the Formula Student Driverles vehicles. I also present, how and to what extent can these approaches perform better with the use of pre-recorded map.
In the next two chapter I review the theoretical background of the used frameworks, which is the SLAM and the convolutional neural networks. I present both graph based and the Bayesian filtering based solutions for the SLAM problem, and the architecture of the Single Shot MultiBox detector as an example for the convolutional neural networks.
I present the used technologies, such as the Android operating system, the ARCore and the Tensorflow frameworks, the train of the used convolutional neural network and the application itself. Finally, the resulting map of the application is compared and evaluated against the map built by the Formula Student Driverless Car of the BME-FRT team (the FSD team of TU Budapest).