Classification of Point Cloud for Road Scene Understanding with Multiscale Voxel Deep Network

Xavier Roynard (1) , Jean-Emmanuel Deschaud (1) , François Goulette (1)

1 CAOR - Centre de Robotique

In this article we describe a new convolutional neural network (CNN) to classify 3D point clouds of urban scenes. Solutions are given to the problems encountered working on scene point clouds, and a network is described that allows for point classification using only the position of points in a multi-scale neighborhood. This network enables the classification of 3D point clouds of road scenes necessary for the creation of maps for autonomous vehicles such as HD-Maps. On the reduced-8 Semantic3D benchmark [Hackel et al., 2017], this network, ranked second, beats the state of the art of point classification methods (those not using an additional regularization step as CRF). Our network has also been tested on a new dataset of labeled urban 3D point clouds for semantic segmentation.

Mots clés: Deep Learning; Point Cloud; Classification; Segmentation

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]; Apprentissage [cs.LG]

Xavier Roynard, Jean-Emmanuel Deschaud, François Goulette. Classification of Point Cloud for Road Scene Understanding with Multiscale Voxel Deep Network. 10th workshop on Planning, Perceptionand Navigation for Intelligent Vehicules PPNIV'2018, Oct 2018, Madrid, Spain. ⟨hal-01763469v2⟩