| Based on the existing studies of intelligent autonomous vehicle perception system-s, combined with multiple sensors fusion, this dissertation focuses on road and lane detection problem, realtime traffic sign detection and recognition problem, realtime in-tersection location problem and independently develops a set of autonomous vehicle perception system. The main dissertation consists of three parts as follows.1. This study proposes a novel real-time optimal-drivable-region and lane detection system for autonomous driving based on multilevel fusion of Light Detection and Ranging (LIDAR) and vision data. Our system successfully handles both struc-tured and unstructured roads. Based on imaging model, this study also presented a multi-lane detection and following algorithm which can efficiently and accurate-ly detect and follow multiple lanes in the complex environments, particularly in the presence of interruption of non-lane road markings.2. This study proposes a multiple features fusion based traffic sign detection method and a ASIFT based traffic sign recognition method. With forbidden signs for ex-ample, through analysis of the inherent characteristics of the traffic signs, making the most of different features for local and global expressing, this study proposes an adaptive weighting serial fusion of features, combining with machine learning method to realize traffic sign detection. Considering the multi-view traffic signs on nature Scenario. This study also proposes an ASIFT-MATCH based multi-view traffic sign recognition method.3. This study proposes an effective real-time intersection detection and recognition approach for autonomous driving in unknown environment based on the point cloud data acquired by a3D laser scanner mounted on the vehicle. The intersec- tion detection and recognition is formulated as a classification problem, namely classification of roads as segments or intersections and subclassification of inter-sections as T-shaped or+-shaped intersections as well. |
PDF Full Text Request