On-street Parking Statistics Using LiDAR Mobile Mapping

This paper presents a procedure to extract parking statistics from a 3D point cloud recorded with two 2D LiDAR sensors mounted on a vehicle. Policy makers can use these parking statistics to reduce parking search traffic in cities by identifying parking characteristics and adjusting current parking rules and policies. The extraction procedure basically consists of an object segmentation and a classification step. For object segmentation, a region growing approach is used to extract the ground surface and to separate distinct objects. For object classification, a random forest classifier is employed with various local and global point features to identify the characteristic shape of vehicles. Comparing the point clouds of both LiDAR scanners allows the exclusion of moving vehicles from the result. A second segmentation in a finer raster after classification is used to reduce the occurrence of undersegmentation. The procedure is evaluated on a 5.5 km track including a residential and a commercial district with parallel and perpendicular parking in a large city in Germany. The results reveal reliable detection of parked vehicles in most situations and therefore approve its suitability for parking studies. Multiple statistics like vehicle dimensions, parking gaps and temporal behavior can be extracted from this procedure. As an example, the occupancy of street segments in the course of one day is presented.