學習三維圖卷積網路於點雲分析

Abstract

在自駕車、無人機、擴增實境等應用當中,電腦如何利用三維資料來感知空間資訊,是至關重要的問題。其中,點雲在三維電腦視覺領域是一種相當普遍且重要的資料型式。然而,不像是二維影像由規則的相素組成,點雲是由一堆點所形成的集合,能有效率地描述物體表面的資訊,卻也因為無規則排列,在分析處理上具有相當大的挑戰。雖然近年來有許多研究團隊及學者,針對分析點雲提出了各種演算法,並也在一些簡單的任務達到了不錯的表現,然而這些演算法皆會在點雲被平移、縮放、旋轉等變化後,準確率會大為下降,這在許多應用當中都是不被允許的。在這篇論文中,我們提出了三維圖卷積網路,該演算法能夠在不同解析度下分析點雲的幾何特徵,經過適當的訓練,能夠進行點雲的分類、語意分割、室內場景辨識。藉由實驗在虛擬或是真實感測到的點雲資料上,我們驗證了三維圖卷積網路能夠用相對輕量化的模型,達到與其他團隊一樣甚至更好的表現,並在各種劇烈的平移、縮放下維持準確率不變。

Point clouds are among the popular geometry representations in 3D vision. However, unlike 2D images with pixel-wise layouts, such representations containing unordered data points which make the processing and understanding the associated semantic information quite challenging. Although a number of previous works attempt to analyze point clouds and achieve promising performances, their performances would degrade significantly when data variations like shift and scale changes are presented. In this thesis, we propose 3D Graph Convolution Networks (3D-GCN), which is designed to extract local 3D features from point clouds across scales, while shift and scale-invariance properties are introduced. The novelty of our 3D-GCN lies in the definition of learnable kernels with a graph max-pooling mechanism. I show that 3D-GCN can be applied to 3D classification and segmentation tasks, with ablation studies and visualizations verifying thedesign of 3D-GCN.