Semantic Segmentation using Convolutional Neural Networks

Abstract

In order for a self-driving system to function well., it must be able to assess the current environment. This new technique relies on precise processing of visual signals in real time. Until recent advancements in deep learning algorithms., such efficiency in processing time and accuracy was not possible because to the complex interplay between pixels in each frame of the incoming camera data. This study provides a feasible approach to perform semantic segmentation for self-driving cars. To create the proposed model., the convolutional neural networks, auto-encoders., and a semantic network design are integrated. To train and evaluate the proposed model, this study makes use of the CamVid dataset, which has undergone an extensive data enhancement. The collected data is then used to verify the proposed model by comparing it to many baseline models found in the literature.