散らかった屋内環境における位相的持続特徴に基づく物体認識
Topologically Persistent Features-based Object Recognition in Cluttered Indoor Environments
物体点群を2Dスライスに分割しパーシステントホモロジーで形状特徴を抽出する新しい位相記述子を提案し、遮蔽物体の認識でDGCNNとSimpleViewを上回った。
著者: Ekta U. Samani, Ashis G. Banerjee
分類: cs.CV, cs.RO
原文アブストラクト
Recognition of occluded objects in unseen indoor environments is a challenging problem for mobile robots. This work proposes a new slicing-based topological descriptor that captures the 3D shape of object point clouds to address this challenge. It yields similarities between the descriptors of the occluded and the corresponding unoccluded objects, enabling object unity-based recognition using a library of trained models. The descriptor is obtained by partitioning an object's point cloud into multiple 2D slices and constructing filtrations (nested sequences of simplicial complexes) on the slices to mimic further slicing of the slices, thereby capturing detailed shapes through persistent homology-generated features. We use nine different sequences of cluttered scenes from a benchmark dataset for performance evaluation. Our method outperforms two state-of-the-art deep learning-based point cloud classification methods, namely, DGCNN and SimpleView.
