Semantic segmentation of images applies in various fields like medical image diagnostics, facial segmentation, satellite image processing, robotics, and autonomous driving vehicles. Semantic image segmentation means labeling image regions with predefined classes of objects represented in that region. Autonomous self-driving vehicles is a trending field of research that requires semantic segmentation of roadside scenes to discover drivable areas. In this research work, we have investigated the semantic segmentation of road scenes captured in unstructured environments. Inspired by deep learning, we have implemented popular deep Convolutional network architectures - FCN8, U-Net, PSPNet, LINKNet, and DEEPLABV3+ for performing semantic segmentation of images using Indian Driving Dataset (IDD) (Lite). The performance of each of these models has been analyzed, among which DEEPLABV3+ produced a better accuracy when compared with the other networks.