DC3DCD: unsupervised learning for multiclass 3D point cloud change detection

TL;DR

This paper proposes an unsupervised method, called DeepCluster 3D Change Detection (DC3DCD), to detect and categorize multiclass changes at point level, and builds upon the DeepCluster approach, originally designed for image classification, to handle complex raw 3D PCs and perform change segmentation task.

Abstract

In a constant evolving world, change detection is of prime importance to keep updated maps. To better sense areas with complex geometry (urban areas in particular), considering 3D data appears to be an interesting alternative to classical 2D images. In this context, 3D point clouds (PCs), whether obtained through LiDAR or photogrammetric techniques, provide valuable information. While recent studies showed the considerable benefit of using deep learning-based methods to detect and characterize changes into raw 3D PCs, these studies rely on large annotated training data to obtain accurate results. The collection of these annotations are tricky and time-consuming. The availability of unsupervised or weakly supervised approaches is then of prime interest. In this paper, we propose an unsupervised method, called DeepCluster 3D Change Detection (DC3DCD), to detect and categorize multiclass changes at point level. We classify our approach in the unsupervised family given the fact that we extract in a completely unsupervised way a number of clusters associated with potential changes. Let us precise that in the end of the process, the user has only to assign a label to each of these clusters to derive the final change map. Our method builds upon the DeepCluster approach, originally designed for image classification, to handle complex raw 3D PCs and perform change segmentation task. An assessment of the method on both simulated and real public dataset is provided. The proposed method allows to outperform fully-supervised traditional machine learning algorithm and to be competitive with fully-supervised deep learning networks applied on rasterization of 3D PCs with a mean of IoU over classes of change of 57.06\% and 66.69\% for the simulated and the real datasets, respectively.

Figures (12)

• Figure 1: Illustration of our proposed method: DC3DCD. It is trained by alternatively clustering deep features to match a pseudo-label to each point of PC 2. These pseudo-labels are used to optimize the back-bone trainable parameters.
• Figure 2: Back-bone architectures used in our experiments.
• Figure 3: User guided mapping of predicted clusters to real classes. For the $K$ predicted clusters, a mapping with the corresponding real class is performed by a user to obtain the final change segmentation of the PC. 5 mappings are provided for the sake of illustration. Segmenting the whole dataset requires $K$ annotations only. This is far less than the millions of points that need to be annotated in order to build training and validation sets in a supervised setting.
• Figure 4: Analysis of the behavior of DC3DCD during the training. The evolution of clustering quality (a) is given thanks to the NMI between the clustering and the real labels of Urb3DCD-V2 dataset. The NMI between the clustering at epoch $t$ and the clustering at epoch $t-1$ gives the cluster reassignment (b).
• Figure 5: Pseudo-cluster entropy at epoch 10 and 50. The entropy is a measure of purity of pseudo-clusters. The lower the entropy, the purer the pseudo-cluster. Pseudo-clusters are sorted in increasing entropy values.