Noise-Resistant Deep Learning for Object Classification in Three-Dimensional Point Clouds Using a Point Pair Descriptor

Abstract

Object retrieval and classification in point cloud data is challenged by noise, irregular sampling density and occlusion. To address this issue, we propose a point pair descriptor that is robust to noise and occlusion and achieves high retrieval accuracy. We further show how the proposed descriptor can be used in a 4D convolutional neural network for the task of object classification. We propose a novel 4D convolutional layer that is able to learn class-specific clusters in the descriptor histograms. Finally, we provide experimental validation on 3 benchmark datasets, which confirms the superiority of the proposed approach.Digital Object Identifier (DOI): see top of this page. Input point cloudPoint pair function values quantized in a 4D histogram 4D to 2D reshape Class … Proposed neural network Descriptor 2D to 1D reshape 1) We present a novel 4D convolutional neural network architecture that takes a 4D descriptor as input and out-

Deep Learning for 3D Point Clouds: A Survey

“…Inspired from the construction of shape context descriptor [89], Xie et al [81] proposed a novel ShapeContextNet architecture by combining affinity point selection and compact feature aggregation into a soft alignment operation using dot-product self-attention [90]. To address noise and occlusion in 3D point clouds, Bobkov et al [91] fed handcrafted point pair function based 4D rotation invariant descriptors into a 4D convolutional neural network. Prokudin et al [92] first randomly sampled a basis point set with a uniform distribution from a unit ball, and then encoded point cloud as minimal distances to the basis point set, which converts the point cloud to a vector with a relatively small fixed length.…”

Novel CE-CBCE feature extraction method for object classification using a low-density LiDAR point cloud

“…Employing machine learning for object classifiers has been a major interest of researchers as a means to train extracted features including for LiDAR point cloud classification [39,40]. Bobkov et al [41] implement a convolutional neural network (CNN) with 5 filters and pooling for layer extraction. Whereas Tian et al [30] implemented multiple object features with annotated labels incorporated with an initialized neural network.…”

Addressing the Sim2Real Gap in Robotic 3-D Object Classification

“…Another direction is to learn from rotation-invariant features as is explored in [4], [30], [3]. While a new representation is introduced in [4], most work take inspiration from classical feature descriptors and explore the potential combination with deep networks.…”

“…While a new representation is introduced in [4], most work take inspiration from classical feature descriptors and explore the potential combination with deep networks. For example, [3] takes inspiration from the Point Pair Features (PPF) descriptor [7] and use convolutional neural networks to learn a multi-dimensional histogram without losing the correlation between the features. The ESF descriptor [31] is another handcrafted descriptor and is designed specifically for classification.…”