Search

With the gradual growth of deep learning in machine vision, efficient extraction of 3D point clouds becomes significant. The raw data of the 3D point cloud are sparse, disordered, and immersed in noise, which makes it difficult to classify and segment. Whether 3D point clouds can be classified and segmented or not, the local feature is an essential ingredient. Therefore, this paper proposes a GateNet-based PointNet++ network (G-PointNet++). G-PointNet++ extracts local features more accurately than PointNet++ by suppressing irrelevant features and emphasizing important features. Meanwhile, it refines the feature adaptively. Besides, the SENet and attention mechanism are introduced into PointNet++. G-PointNet++ was evaluated on the public ModelNet dataset, ShapeNet dataset, and S3DIS dataset, and its effectiveness in classification and segmentation tasks was verified. In the classification task, G-PointNet++ achieves an overall classification accuracy (OA) of 95.5% on ModelNet10 and 93.3% on ModelNet40. In the segmentation task, the mIoU of G-PointNet++ reaches 85.5% on ShapeNet. These experiments show that G-PointNet++ achieves better performance and saves more time than PointNet++, and its overall accuracy is higher than that of PCT network on ModeNet40. [ABSTRACT FROM AUTHOR]

Recognizing 3D objects based on local feature descriptors, in point cloud scenes with occlusion and clutter, is a very challenging task. Most existing 3D local feature descriptors rely on normal information to encode local features, however, they ignore the normal-sign-ambiguity issue, which greatly limits their descriptiveness and robustness. This paper proposes a method called VOxelization in Invariant Distance space for 3D object recognition. First, we propose a VOID descriptor that is invariant to normal-sign-ambiguity, and is also rotation-invariant, distinctive, robust, and efficient. Second, a VOID-based 3D object recognition method considering the self-similarity between local features is proposed to enhance the recognition performance. Five standard datasets are employed to validate our proposed method as well as comparison with the state-of-the-arts. The results suggest that: (1) VOID descriptor is invariant to normal-sign-ambiguity, distinctive, and robust; (2) VOID-based 3D object recognition achieves outstanding recognition performance, i.e., 99.47%, 93.07% and 99.18%, on the U3OR, Queen's and Ca' Foscari Venezia datasets, respectively. [ABSTRACT FROM AUTHOR]

In a large number of real-life scenes and practical applications, 3D object detection is playing an increasingly important role. We need to estimate the position and direction of the 3D object in the real scene to complete the 3D object detection task. In this paper, we propose a new network architecture based on VoteNet to detect 3D point cloud targets. On the one hand, we use channel and spatial dual-domain attention module to enhance the features of the object to be detected while suppressing other useless features. On the other hand, the SIFT operator has scale invariance and the ability to resist occlusion and background interference. The PointSIFT module we use can capture information in different directions of point cloud in space, and is robust to shapes of different proportions, so as to better detect objects that are partially occluded. Our method is evaluated on the SUN-RGBD and ScanNet datasets of indoor scenes. The experimental results show that our method has better performance than VoteNet. [ABSTRACT FROM AUTHOR]

Accurate online recognition of raw material particle size is crucial for the green and efficient operation of blast furnaces. In this study, a high-resolution image acquisition system for blast furnace raw materials was established, and multiple batches of samples were collected for sintered ore and coke particle size analysis. To improve the robustness and generalization capability of the model, image enhancement techniques such as translation, scaling, cropping, random rotation, and brightness adjustment were applied to obtain the basic image datasets for sintered ore and coke. Among four mainstream instance segmentation algorithms, YOLACT++ was selected as the base network. To enhance the recognition accuracy of raw material similarities and subtle features, improvements were made to the YOLACT++ backbone network by replacing the original ResNet101 with DetNet59, and a composite loss function was constructed. Based on the enhanced YOLACT++ algorithm, a particle size recognition model was developed, achieving accuracy rates (PA), recall rates (Recall), F1 scores, and mean intersection over union (MIoU) all exceeding 90%. To address issues such as misidentification of sintered ore voids and over-segmentation in the particle size recognition model, an optimized 3D point cloud concavity and convexity segmentation algorithm was employed. By re-inputting the segmented 3D point cloud results into the particle size recognition model, problems such as void misidentification and over-segmentation were effectively resolved, achieving the multimodal fusion of deep learning and 3D point clouds and improving the accuracy and stability of sintered ore particle size recognition. [ABSTRACT FROM AUTHOR]

Point cloud completion represents a complex task that entails predicting the complete geometry of a 3D shape from a set representation of the partial observational points. This paper presents a novel point cloud completion method, called Cascaded Graph Convolutional Completion Network (CGCN). Our method leverages a cascaded encoder–decoder architecture to predict the missing points from the input, the partial points. These predicted missing points are then concatenated with the input points to form the complete shape. Our architecture consists of two main modules: a Multi-Level Feature Extraction Encoder (MFE) and a Folding-Refinement Decoder (FRD). The MFE is composed of the Edge Feature Extraction Module (EFE) and the Global Feature Extraction Module (GFE). The encoder initiates by using several EFEs to secure multi-level local features through the execution of graph convolutional operations on each point. It then employs a GFE to extract global features from the final level features. Consequently, the MFE aggregates features containing local information into global features, offering an output that carries both local and global features. Our FRD contains both a folding block and several refinement blocks. Initially, the folding block morphs 2D grids with features output by the MFE to predict the initial missing points. The decoder then iteratively applies a series of refinement blocks to refine these initial missing points and eventually obtain the output of the predicted missing points. The FRD merges multi-level features and global features to deliver the shape of the predicted missing part. To achieve the complete shape, we integrate the predicted missing part, derived from the FRD and the input points. We then sample this unified point cloud to secure the final output. The effectiveness and competitiveness of our model are validated through experiments on the ShapeNet dataset, using Chamfer Distance (CD) as metrics. [ABSTRACT FROM AUTHOR]

Motivated by how humans perceive geometry and color to recognize objects, we propose a boundary embedded attentional convolution (BEACon) network for point cloud instance segmentation. At the core of BEACon, we introduce the attentional weight in the convolution layer to adjust the neighboring features, with the weight being adapted to the relationship between geometry and color changes. As a result, BEACon makes use of both geometry and color information, takes instance boundary as an important feature, and thus learns a more discriminative feature representation in the neighborhood. Experimental results show that BEACon outperforms the state-of-the-art by a large margin. Ablation studies are also provided to prove the large benefit of incorporating both geometry and color into attention weight for instance segmentation. [ABSTRACT FROM AUTHOR]