Recognition of wave-dissipating blocks with multiple types from 3D large-scale point clouds

抄録

Recognition of wave-dissipating blocks stacked on the coast is a significant and challenging task. This task consists of semantic segmentation, instance segmentation, and precise 6D pose estimation. The convolutional neural network (CNN) trained by the block fall simulation is used to classify the original point cloud with multiple types into the single one and to partition into the subsets corresponding both to the individual blocks. A descriptor-based method is then used for the 6D pose estimation of the block. We evaluated our method on several scenes containing wave-clipping blocks with multiple block types. The experimental results show that our method is effective and efficient.