Point Cloud 3D Object Detection Based on Multi-Scale Features and Grouped Convolutions

Point cloud-based three-dimensional (3D) object detection is a critical task in autonomous driving, robotics, and augmented reality, where accurate localization and classification of objects are essential under diverse and challenging scenarios. This work introduces a point cloud-based multi-target 3D object detection framework using Light Detection and Ranging (LiDAR) sensors. The key contribution lies in improving the encoding technique to preserve spatial information and adapt to varying point densities, enabling efficient processing of raw LiDAR data for robust 3D detection. Specifically, we design an efficient network architecture that incorporates a single unified classification head to jointly handle positive and negative samples, simplifying the design and improving training stability. Our model detects all classes (car, pedestrian, cyclist) simultaneously within a single network, enhancing computational efficiency while avoiding separate networks for each class. Furthermore, the proposed feature extraction pipeline combines a voxel-based encoder with a sparsely embedded convolutional backbone and a feature pyramid network, facilitating multi-scale feature representation and effective detection of objects at different scales. This encoding method facilitates the efficient processing of raw LiDAR data, enabling accurate object detection across diverse scenarios. As a result, our model achieves state-of-the-art performance on the KITTI dataset, surpassing baseline methods such as PointPillars and VoxelNet by delivering superior Average Precision (AP) across all difficulty levels for the car class in both bird’s-eye view (BEV) and 3D object detection tasks. Specifically, it achieves AP40 scores of 92.84%, 88.63%, and 85.68% in BEV detection and 87.84%, 76.65%, and 73.60% in 3D detection for Easy, Moderate, and Hard levels, respectively. These results highlight the effectiveness of our encoding technique in enhancing model efficiency and detection accuracy across diverse scenarios.

Nowadays, point clouds are frequently gathered by 3D scanners such as Lidar and Kinect, which produces thousands of point cloud models. Point cloud processing is vital to 3D vision, especially 3D object recognition, positioning, and navigation technology. Addressing uneven data density caused by coordinate frame transformations and the inherent problem of insufficient context connection in point clouds, the DiM-PCNet (Multi-scale and Multi-level Point Clouds Classification Network, and the Di is a prefix to represent double M in Multi-scale and Multi-level) is proposed in this paper. DiM-PCNet is provided for object classification with multi-scale and multi-level features. We encode the point cloud in multi-scale and fully fuse the features with the raw point cloud for keeping the context relationship. In DiM-PCNet, we sample the point clouds from eight parts for multi-scales feature extraction. The multi-scales features are fully fused by multi-level pyramid models. The multi-scale and multi-level strategies are applied in DiM-PCNet, in which the abundant and important features of point clouds are extracted and utilized in the 3D object classification. It is worth noting that the DiM-PCNet feature block can be embedded into the segmentation net, where the accuracy achieved is 87.1%. We conducted experiments on ShapeNet and ModelNet40 and the experimental results show that DiM-PCNet achieves state-of-the-art performance in 3D object classification. The experiment shows competitive performance on robustness and segmentation tasks.

The effective use of multi-scale features remains an open problem for object detection tasks. Recently, proposed object detectors have usually used Feature Pyramid Networks (FPN) to fuse multi-scale features. Since Feature Pyramid Networks use a relatively simple feature map fusion approach, it can lead to the loss or misalignment of semantic information in the fusion process. Several works have demonstrated that using a bottom-up structure in a Feature Pyramid Network can shorten the information path between lower layers and the topmost feature, allowing an adequate exchange of semantic information from different layers. We further enhance the bottom-up path by proposing a multi-scale residual aggregation Feature Pyramid Network (MSRA-FPN), which uses a unidirectional cross-layer residual module to aggregate features from multiple layers bottom-up in a triangular structure to the topmost layer. In addition, we introduce a Residual Squeeze and Excitation Module to mitigate the aliasing effects that occur when features from different layers are aggregated. MSRA-FPN enhances the semantic information of the high-level feature maps, mitigates the information decay during feature fusion, and enhances the detection capability of the model for large objects. It is experimentally demonstrated that our proposed MSRA-FPN improves the performance of the three baseline models by 0.5–1.9% on the PASCAL VOC dataset and is also quite competitive with other state-of-the-art FPN methods. On the MS COCO dataset, our proposed method can also improve the performance of the baseline model by 0.8% and the baseline model’s performance for large object detection by 1.8%. To further validate the effectiveness of MSRA-FPN for large object detection, we constructed the Thangka Figure Dataset and conducted comparative experiments. It is experimentally demonstrated that our proposed method improves the performance of the baseline model by 2.9–4.7% on this dataset and can reach up to 71.2%.

Effectively capturing the multi-scale, multi-channel, and boundary features of target images is beneficial for improving the accuracy of salient object detection. A multichannel feature enhancement module (MFEM) is constructed by concatenating two feature maps along the channel, performing convolution activation, dot product addition, and convolution operations. An edge feature enhancement module (EFEM) is constructed by performing an inverse attention mechanism and Gaussian-Laplace transform on the predicted map, channel-wise concatenation, and convolution operations with the feature submap. Multi-scale features are extracted using a SwinTransformer encoder and enhanced using the MFEM. The enhanced features are then fused using a cross-attention mechanism for multi-level features. The fused feature map, the high-frequency feature map obtained by Gaussian-Laplace transforming the predicted map, and the external edge features are input into the EFEM to obtain the salient object detection results. This results in a novel salient object detection model with multi-scale, multi-channel, and boundary feature enhancement: MfENet. Experimental results show that the Mean Absolute Error (MAE), S-measure and F-measure accuracies of our method on the HKU-IS dataset are 0.018,0.924 and 0.959 respectively. These results demonstrate the effectiveness of our method. Furthermore, this work highlights the importance of multi-scale and boundary awareness in enhancing salient object detection, and its contribution to the technical field relevant to the theme of this conference.

Multi-level refinement enriched feature pyramid network for object detection

We propose octree-based transformers, named OctFormer, for 3D point cloud learning. OctFormer can not only serve as a general and effective backbone for 3D point cloud segmentation and object detection but also have linear complexity and is scalable for large-scale point clouds. The key challenge in applying transformers to point clouds is reducing the quadratic, thus overwhelming, computation complexity of attentions. To combat this issue, several works divide point clouds into non-overlapping windows and constrain attentions in each local window. However, the point number in each window varies greatly, impeding the efficient execution on GPU. Observing that attentions are robust to the shapes of local windows, we propose a novel octree attention, which leverages sorted shuffled keys of octrees to partition point clouds into local windows containing a fixed number of points while permitting shapes of windows to change freely. And we also introduce dilated octree attention to expand the receptive field further. Our octree attention can be implemented in 10 lines of code with open-sourced libraries and runs 17 times faster than other point cloud attentions when the point number exceeds 200 k. Built upon the octree attention, OctFormer can be easily scaled up and achieves state-of-the-art performances on a series of 3D semantic segmentation and 3D object detection benchmarks, surpassing previous sparse-voxel-based CNNs and point cloud transformers in terms of both efficiency and effectiveness. Notably, on the challenging ScanNet200 dataset, OctFormer outperforms sparse-voxel-based CNNs by 7.3 in mIoU. Our code and trained models are available at https://wang-ps.github.io/octformer.

Current methods used in the field of 3D target detection generally have the problem that target context features are not rich enough to achieve accurate dynamic multi-target detection in feature extraction. A 3D dynamic object detection algorithm based on voxel point cloud fusion is proposed. The algorithm uses a two-stage detection architecture of multiple and fusion of multi-scale feature. In the first stage, point cloud is directly processed to extract key point features and the voxel space is divided to extract multi-scale voxel features, and two features are firstly fused to generate preselection frame. In the second stage, a reference point is set in each voxel and surrounding key points are absorbed for the second feature fusion, and the final feature is input to a detection module to achieve optimization of the preselection box. Additionally, for the problem of inconsistency between classification and location reliability, a mandatory consistency loss function is proposed to further improve accuracy of detection. Compared with other algorithms in Kitti, Waymo and nuScene datasets, the accuracy of 3D dynamic target detection is 92.10%, the results on the real vehicle platform show that the algorithm has strong robustness, portability and generalization ability.

Aiming at the problem that the relationship between geometric features and semantic features is ignored in the point cloud data downsampling process, which leads to inaccurate segmentation of object boundaries and structural details, this paper proposes a 3D point cloud semantic segmentation network based on multi-scale dense nested type. Firstly, a dense nested network architecture is constructed by nesting multiple multi-scale feature fusion modules to fuse multi-scale features of different directions between encoder-decoder paths, so as to effectively propagate local ge-ometric context information and enhance the ability of cross-scale information interaction. Secondly, a local feature aggregation unit is constructed in the multi-scale feature fusion module, which strengthens the structural awareness within the local point set based on graph convolution and attention mechanism, and promotes the complementarity of local geometric features and abstract semantic information. Then, the cross-layer multi-loss supervision module is combined to further optimize the multi-scale feature propagation, which makes the network training more stable and improves the point cloud segmentation accuracy. Finally, this paper verified the proposed network on the S3DIS dataset. The experimental results show that the proposed network has the mean intersection over Union of 71.2% and the overall accuracy of 88.7%, which is 1.2 and 0.7 percentage points higher than RandLA-Net, respectively, which proves that the proposed network can effectively improve the accuracy of 3D point cloud semantic segmentation.

Compared with the existing modes of LiDAR, single-photon LiDAR (SPL) can acquire terrain data more efficiently. However, influenced by the photon-sensitive detectors, the collected point cloud data contain a large number of noisy points. Most of the existing denoising techniques are based on the sparsity assumption of point cloud noise, which does not hold for SPL point clouds, so the existing denoising methods cannot effectively remove the noisy points from SPL point clouds. To solve the above problems, we proposed a novel multistage denoising strategy with fused multiscale features. The multiscale features were fused to enrich contextual information of the point cloud at different scales. In addition, we utilized multistage denoising to solve the problem that a single-round denoising could not effectively remove enough noise points in some areas. Interestingly, the multiscale features also prevent an increase in false-alarm ratio during multistage denoising. The experimental results indicate that the proposed denoising approach achieved 97.58%, 99.59%, 95.70%, and 77.92% F1-scores in the urban, suburban, mountain, and water areas, respectively, and it outperformed the existing denoising methods such as Statistical Outlier Removal. The proposed approach significantly improved the denoising precision of airborne point clouds from single-photon LiDAR, especially in water areas and dense urban areas.

Registering 3D point clouds with low overlap is challenging in 3D computer vision, primarily due to difficulties in identifying small overlap regions and removing correspondence outliers. We observe that the neighborhood similarity can be utilized to detect point correspondence, and the consistent neighborhood correspondence can be used as a criterion to detect robust overlapping regions. So that a Double-layer Multi-scale Star-graph (DMS) structure is proposed to detect robust correspondences using two different types of multi-scale star-graphs. The first-layer Multi-scale Neighbor Feature Star-graphs (MNFS) takes each point as the center and its multi-scale nearest neighbors as the leaves. The MNFS enables to establish the initial correspondence candidate set between the two point clouds based on multi-scale neighborhood topology and feature similarity. Subsequently, each pair of corresponding points find their nearest neighbors within the correspondence sets to construct a Multi-scale Matching Star-graphs (MMS) on each side, so the mutual correspondence relationships between the MMS vertices are identified. These identified mutual correspondences are treated as vertices to construct the Multi-scale Correspondence Star-graphs (MCS), that indicate the relationships among the correspondences. We design edge weight and vertex weight criterion in MCS to detect only the robust correspondence set that has strong neighborhood consistency, so as to reject the outliers. Finally, the point cloud registration is conducted based on the detected robust correspondence. The experimental results demonstrate clearly that the proposed DMS method exhibits superior robustness when compared to existing state-of-the-art registration algorithms.

Aiming at classifying the feature of LiDAR point cloud data in complex scenario, this paper proposed a deep neural network model based on multi-scale features and PointNet. The method improves the local feature of PointNet and realize automatic classification of LiDAR point cloud under the complex scene. Firstly, this paper adds multi-scale network on the basis of PointNet network to extract the local features of points. And then these local features of different scales are composed into a multi-dimensional feature through the fully connected layer, and combined with the global features extracted by PointNet, the scores of each point class are returned to complete the point cloud classification. The deep neural network model proposed in this paper is verified using the Semantic3D dataset and the Vaihingen dataset provided by ISPRS. The experimental results show that the proposed algorithm achieves higher classification accuracy compared with other neural networks used for point cloud classification.

HRNet: 3D object detection network for point cloud with hierarchical refinement

Accurate object detection is a fundamental requirement for autonomous systems to operate in dynamic urban environments. Considering the intricacy of the environment and the occlusion of objects, various point cloud based three-dimensional (3D) object detection methods have been proposed, such as point-based or voxel-based methods. In this paper, a feature selection mechanism is proposed in a voxel-based method to generate bird’s eye view (BEV) feature maps from the original point cloud. For the 3D Backbone, the U-shaped structure and single scale feature selection module are combined. After combining high-level semantics and low-level fine-grained features, the optimized features after the BEV feature maps are applied to region of interest (RoI) refinement, so that the voxel features can better serve the subsequent 3D object detection. The experimental results on KITTI dataset show higher 3D object detection accuracy compared to the state-of-the-art 3D detection methods, which reflects the effectiveness of the proposed architecture.

The optical remote sensing images (RSIs) show various spatial resolutions and cluttered background, where salient objects with different scales, types, and orientations are presented in diverse RSI scenes. Therefore, it is inappropriate to directly extend cutting-edge saliency detection methods for conventional RGB images to optical RSIs. Besides, the existing saliency models targeting RSIs often render imperfect saliency maps, where some of them are with coarse boundary details. To solve this problem, this article attempts to introduce the edge information to precisely detect salient objects in RSIs. Accordingly, we propose an edge-aware multiscale feature integration network (EMFI-Net) for salient object detection by conducting multiscale feature integration under the explicit and implicit assistance of salient edge cues. Specifically, our network contains two parts including the encoder and decoder. First, the encoder extracts multiscale deep features from three RSIs with different resolutions, where the high-level deep semantic features from three RSIs are integrated using a cascaded feature fusion module. Second, the encoder explicitly enriches the multiscale deep features by integrating the salient edge cues extracted by a salient edge extraction module. Meanwhile, we also implicitly deploy an edge-aware constraint to the supervision of the saliency map prediction by introducing a hybrid loss function. Finally, the decoder integrates the enriched multiscale deep features in a coarse-to-fine way, yielding a high-quality saliency map. The experiments conducted on two public optical RSI datasets clearly prove the effectiveness and superiority of the proposed EMFI-Net against the state-of-the-art saliency models.

3D object detection from raw and sparse point clouds has been far less treated to date, compared with its 2D counterpart. In this paper, we propose a novel framework called FVNet for 3D front-view proposal generation and object detection from point clouds. It consists of two stages: generation of front-view proposals and estimation of 3D bounding box parameters. Instead of generating proposals from camera images or bird's-eye-view maps, we first project point clouds onto a cylindrical surface to generate front-view feature maps which retains rich information. We then introduce a proposal generation network to predict 3D region proposals from the generated maps and further extrude objects of interest from the whole point cloud. Finally, we present another network to extract the point-wise features from the extruded object points and regress the final 3D bounding box parameters in the canonical coordinates. Our framework achieves real-time performance with 12ms per point cloud sample. Extensive experiments on the 3D detection benchmark KITTI show that the proposed architecture outperforms state-of-the-art techniques which take either camera images or point clouds as input, in terms of accuracy and inference time.