Deep Learning-Based Semantic Segmentation in Autonomous Driving

Abstract

Perception is the first and most important task of any autonomous driving system. It extracts visual information about the surrounding environment of the vehicle. The perception data is then fed to a decision-making system to provide the optimum decision given a specific scenario to avoid potential collisions. In this paper, we have developed variants of the U-Net model to perform semantic segmentation on urban scene images to understand the surroundings of an autonomous vehicle. The U-N et model and its variants are adopted for semantic segmentation in this project to account for the power of the UNet in handling large and small datasets. We have also compared the best-performing variant with other commonly used semantic segmentation models. The comparative analysis was performed using three well-known models, including FCN-16, FCN-8, and SegNet. After conducting sensitivity and comparative analysis, it is concluded that the U-Net variants performed the best in terms of the Intersection over Union (IoU) evaluation metric and other quality metrics.