APPLICABILITY OF NEURAL NETWORKS FOR IMAGE CLASSIFICATION ON OBJECT DETECTION IN MOBILE MAPPING 3D POINT CLOUDS
In this work, we present an approach that uses an established image recognition convolutional neural network for the semantic classification of two-dimensional objects found in mobile mapping 3D point cloud scans of road environments, namely manhole covers and road markings. We show that the approach is capable of classifying these objects and that it can efficiently be applied on large datasets. Top-down view images from the point cloud are rendered and classified by a U-Net implementation. The results are integrated into the point cloud by setting an additional semantic attribute. Shape files can be computed from the classified points.