A Performance Evaluation of Recent Lane Detection Approaches for Intelligent Vehicles in Real Urban Scenarios

Lane detection is a key feature of today’s modern intelligent vehicles and it is especially important for Advanced Driver Assitance systems. Lane detection task is in charge of identifying drivable zones on the road by detecting the lane markers and developing information to describe the road shape. As a result, it has been an active subject of research for decades with significant advances made in recent years. Lane detection methods are typically limited to simple road conditions, making them unsuitable for scenarios involving complicated urban road scenarios. The researchers have been limited in their capacity to test the algorithms since there are not enough rigorous dataset that include a wide variety of data. When it comes to real-world applications of autonomous driving, there are still significant gaps on lane detection task because of the challenging scenarios including different weather conditions, occluded lane lines, lightning conditions and road surfaces that vary cracking, rutting insufficiently removed markings, intense shadow conditions which all contribute to a dangerous driving environment. The purpose of creating an urban scenarios dataset is to identify these gaps and recommend research directions that could be used to bridge the gaps that were identified. The purpose of this thesis is to conduct an exhaustive analysis of the existing literature on lane detection and to identify knowledge gaps for the purpose of further investigation with newly generated dataset. It is hoped that this study will serve as a useful resource for academics in the future who are planning to create lane detection and tracking algorithms for forthcoming autonomous cars in very challenging road scenarios.