Semantic segmentation for autonomous navigation

Autonomna vozila su vozila svjesna vlastite okoline i sposobna samostalnog kretanja u njoj. To im omogućuju razni senzori i algoritmi koji obrađuju njihove podatke. Semantička segmentacija je ključan algoritam učenja autonomnih vozila temeljen na konvolucijskim neuronskim mrežama. Konvolucijske neuronske mreže su posebna vrsta dubokih neuroskih mreža temeljene na prepoznavanju slika. Kako bi se algoritam prilagodio stvarnom svijetu koriste se okruženja koja simuliraju stvarne uvjete okoline. Kada algoritam prođe sve uvjete dane u virtualnom okruženju, tek onda se razmatra testiranje u stvarnoj okolini. CARLA je okruženje koje se koristilo u ovom radu. Rad detaljno opisuje konvolucijske neuronske mreže i njihove prednosti nad navigacijom autonomnih vozila.Autonomous vehicles are vehicles aware of their own environment and capable of independent movement in it. This is made possible by various sensors and algorithms that process their data. Semantic segmentation is a key learning algorithm for autonomous vehicles based on convolutional neural networks. Convolutional neural networks are a special type of deep neural networks based on image recognition. In order to adapt, the algorithm uses simulation of real-world conditions. When the algorithm passes all the conditions given in the virtual environment, only then is testing in the real environment considered. CARLA is the environment used in this paper. The paper describes in detail convolutional neural networks and their advantages over autonomous vehicle navigation