Intersection detection and recognition for autonomous urban driving using a virtual cylindrical scanner

In this study, the authors propose an effective real-time approach for intersection detection and recognition during autonomous driving in an unknown urban environment. The authors approach use point cloud data acquired by a three-dimensional laser scanner mounted on the vehicle. Intersection detection and recognition are formulated as a classification problem whereby roads are classified as segments or intersections and intersections are subclassified as T-shaped or +-shaped. They first construct a novel model called a virtual cylindrical scanner for efficient feature-level representation of the point cloud data. Then they use support vector machine classifiers to resolve the classification problem according to the features extracted. A series of experiments on real-world data sets and in a simulation environment demonstrate the effectiveness and robustness of the authors approach, even in highly dynamic urban environment. They also performed simulation experiments to investigate effects of several critical factors on their proposed approach, such as other vehicles on the road and the advance detection distance.