Automated extraction of manhole covers using mobile LiDAR data

Our work addresses the problem of accurately extracting manhole covers, one specific type of road fixture, from a large volume of mobile light detection and ranging (LiDAR) data. We propose an efficient, step-wise manhole cover extraction method. First, road surface points are extracted and interpolated to generate two-dimensional geo-referenced feature (GRF) images. Next, the method segments manhole cover candidates by applying distance-dependent thresholding and multi-scale tensor voting to the GRF images. Finally, distance-based clustering and a morphological operation are used to extract manhole covers. Experiments were carried out to qualitatively and quantitatively demonstrate the performance of our manhole cover extraction method.