Semantic Segmentation Algorithm for Autonomous Driving using UNET Architectures: A Comparative Study

Abstract In autonomous driving systems, the semantic segmentation task involves scene partition into numerous expressive portions by classifying and labeling every image pixel for semantics. The algorithm used for semantic-segmentation has a vital role in autonomous driving architecture. The main contribution of this paper is the design of UNET architecture for the semantic-segmentation algorithm of autonomous driving. We address semantic image segmentation by the images captured using the open-source Car Learning to Act (CARLA) simulator platform for this research work. Semantic segmentation processes images from the camera obtained from the simulated vehicle for classification into multiple classes, including roads, cars, bikes, lanes, bicycles, trees, sky, etc. This study is intended to capture the merits of the U-NET semantic segmentation Algorithms for autonomous driving using the Keras framework. The proposed algorithm was evaluated on a dataset generated using CARLA simulation software and the CAMViD dataset for different metrics.