Perth-WA Localization Dataset in 3D Point Cloud Maps

Another major contribution of this work is Perth-WA dataset that provides 6DoF annotations for localization. The data comprises a LiDAR map of 4km square region of Perth Central Business District (CBD) in Western Australia. The scenes contain commercial structures, residential areas, food streets, complex routes, and hospital building etc. The data was collected in three different two-hour sessions under day/night conditions with sunny and cloudy weather.  Unlike the existing related dataset, Apollo-SouthBay and Oxford Radar RobotCar Dataset, Perth-WA dataset annotations do not rely on Inertial Measurement Unit (IMU). Instead, the labeling comes directly from the LiDAR frames themselves. To extract the ground-truth poses for Perth-WA dataset, we exploit the map creation process itself. Within a loop, a moving LiDAR frame is registered with a static point cloud, which generates a transformation matrix for the moving frame. To compute the transformation matrix of a frame, we multiply the previous frame's transformation to the transformation matrix of that frame