VeloRegistration based intersection detection for autonomous driving in challenging urban scenarios

In this paper, an improved intersection detection method is proposed by applying the VCS-based algorithm on the registered scans instead of the single scan. Both the registration and intersection detection approach are independent on Global Positioning System (GPS), Geographic Information System (GIS), Inertial Navigation System (INS) or other auxiliaries which have been extensively used in autonomous navigation. The novel registration method named VeloRegistration addresses the data registration problem and moving object detection simultaneously by introducing a tracking-classification operator into iterative matching process. The accurate and massive information generated by VeloRegistration enables us to cope with much more complicated intersection scenarios, especially there are dead zones caused by severe occlusion. Experimental validation including three real world data sets acquired in Wuhan, China substantiate the effectiveness and robustness of our approach in challenging urban scenarios.