Following the leader using a tracking system based on pre-trained deep neural networks
Filipe Wall Mutz, Vinícius dos Santos Cardoso, Thomas Teixeira, Luan F. R. Jesus, Michael A. Golcalves, Rânik Guidolini, Josias Oliveira, Claudine Santos Badue, Alberto Ferreira De Souza · 2017
In this work, we present a software architecture to solve, at some level, the follow the leader problem. This problem consists of an autonomous vehicle trying to track and follow a leader vehicle. To track the leader position in consecutive camera images, we employed the Generic Object Tracking Using Regression Networks (GOTURN). GOTURN is a pre-trained Deep Neural Network capable of tracking generic objects, without application-specific training or fine-tuning. The proposed software architecture was evaluated using a real autonomous vehicle, in four stretches of a University ring road. In all experiments, the autonomous vehicle was able to follow the leader's path with maximum root mean square error of 0.28m.