Detection and Segmentation of 3D Objects in Urban Environments Using Indexation

A procedure for automobile detection on 3D point clouds of urban areas is presented in this work. Point clouds are obtained using an HDL-64E Velodyne LIDAR. The work is divided into two sections: Segmentation, in which the base plane (floor) and its perpendicular planes are extracted using Hough's technique. Next every other object is segmented using MeanShift method; and Indexation, in which all segmented objects are modeled according to a normal direction so that its histograms can be obtained and compared to a pre-loaded histogram database. The reconstructed environment is considered to be semi-structured, meaning that it can be modeled using planes. In the process ROC analysis is used for thresholds optimization.