Title:A Methodological Review of Visual Road Recognition Procedures for Autonomous Driving Applications

Abstract:The current research interest in autonomous driving is growing at a rapid pace, attracting great investments from both the academic and corporate sectors. In order for vehicles to be fully autonomous, it is imperative that the driver assistance system is adapt in road and lane keeping. In this paper, we present a methodological review of techniques with a focus on visual road detection and recognition. We adopt a pragmatic outlook in presenting this review, whereby the procedures of road recognition is emphasised with respect to its practical implementations. The contribution of this review hence covers the topic in two parts -- the first part describes the methodological approach to conventional road detection, which covers the algorithms and approaches involved to classify and segregate roads from non-road regions; and the other part focuses on recent state-of-the-art machine learning techniques that are applied to visual road recognition, with an emphasis on methods that incorporate convolutional neural networks and semantic segmentation. A subsequent overview of recent implementations in the commercial sector is also presented, along with some recent research works pertaining to road detections.