3D Object Detection Research Articles

VIDF-Net: A Voxel-Image Dynamic Fusion method for 3D object detection

In recent years, multi-modal fusion methods have shown excellent performance in the field of 3D object detection, which select the voxel centers and globally fuse with image features across the scene. However, these approaches exist two issues. First, The distribution of voxel density is highly heterogeneous due to the discrete volumes. Additionally, there are significant differences in the features between images and point clouds. Global fusion does not take into account the correspondence between these two modalities, which leads to the insufficient fusion. In this paper, we propose a new multi-modal fusion method named Voxel-Image Dynamic Fusion (VIDF). Specifically, VIDF-Net is composed of the Voxel Centroid Mapping module (VCM) and the Deformable Attention Fusion module (DAF). The Voxel Centroid Mapping module is used to calculate the centroid of voxel features and map them onto the image plane, which can locate the position of voxel features more effectively. We then use the Deformable Attention Fusion module to dynamically calculates the offset of each voxel centroid from the image position and combine these two modalities. Furthermore, we propose Region Proposal Network with Channel-Spatial Aggregate to combine channel and spatial attention maps for improved multi-scale feature interaction. We conduct extensive experiments on the KITTI dataset to demonstrate the outstanding performance of proposed VIDF network. In particular, significant improvements have been observed in the Hard categories of Cars and Pedestrians, which shows the significant effectiveness of our approach in dealing with complex scenarios.

Spatial awareness enhancement based single-stage anchor-free 3D object detection for autonomous driving

The real-time and accurate detection of three-dimensional (3D) objects based on LiDAR is a focal problem in the field of autonomous driving environment perception. Compared to two-stage and anchor-based 3D object detection methods that suffer from inference latency challenges, single-stage anchor-free 3D object detection approaches are more suitable for deployment in autonomous driving vehicles with the strict real-time requirement. However, they face the issue of insufficient spatial awareness, which can result in detection errors such as false positives and false negatives, thereby increasing the potential risks of autonomous driving. In response to this, we focus on enhancing the spatial awareness of CenterPoint, a widely used single-stage anchor-free 3D object detector in the industry. Considering the limited allocation of computational resources and the performance bottleneck caused by pillar encoder, we propose an efficient SSDCM backbone to strengthen feature representation and extraction. Furthermore, a simple BGC neck is devised to weight and exchange contextual information in order to deeply fuse multi-scale features. Combining improved backbone and neck networks, we construct a single-stage anchor-free 3D object detection model with spatial awareness enhancement, named CenterPoint-Spatial Awareness Enhancement (CenterPoint-SAE). We evaluate CenterPoint-SAE on two large-scale and challenging autonomous driving datasets, nuScenes and Waymo. It achieves 53.3% mAP and 62.5% NDS on nuScenes detection benchmark, and runs inference at a speed of 11.1 FPS. Compared to the baseline, the upgraded networks deliver a performance improvement of 1.6% mAP and 1.2% NDS at minor cost. Notably, on Waymo dataset, our method achieves competitive detection performance compared to two-stage and point-based methods.

PesRec: A parametric estimation method for indoor semantic scene reconstruction from a single image

Reconstructing semantic indoor scenes is a challenging task in augmented and virtual reality. The quality of scene reconstruction is limited by the complexity and occlusion of indoor scenes. This is due to the difficulty in estimating the spatial structure of the scene and insufficient learning for object location inference. To address these challenges, we have developed PesRec, an end-to-end multi-task scene reconstruction network for parameterizing indoor semantic information. PesRec incorporates a newly designed spatial layout estimator and a 3D object detector to effectively learn scene parameter features from RGB images. We modify an object mesh generator to enhance the robustness of reconstructing indoor occluded objects through point cloud optimization in PesRec. Using the analyzed scene parameters and spatial structure, the proposed PesRec reconstructs an indoor scene by placing object meshes scaled to 3D detection boxes in an estimated layout cuboid. Extensive experiments on two benchmark datasets demonstrate that PesRec performs exceptionally well for object reconstruction with an average chamfer distance of 5.24 × 10-3 on the Pix3D dataset including 53.61 % mAP for 3D object detection and 79.7 % 3D IoU for the estimation of layout on the commonly-used SUN RGB-D datasets. The proposed computing network breaks through the limitations caused by complex indoor scenes and occlusions, showing optimization results that improve the quality of reconstruction in the fields of augmented reality and virtual reality.

VFL3D: A Single-Stage Fine-Grained Lightweight Point Cloud 3D Object Detection Algorithm Based on Voxels

VFL3D: A Single-Stage Fine-Grained Lightweight Point Cloud 3D Object Detection Algorithm Based on Voxels

Small Object Detection with lightweight PointNet Based on Attention Mechanisms

Abstract In response to the issue of subpar small object detection outcomes, the improved 3D object detection method for Point Net is proposed. First, multi-scale fusion method is added to the channel attention mechanism into the feature extraction stage of Point Net; then, the loss function is adjusted to optimize the backbone network and reduce the number of model parameters while improving the accuracy. The experimental results show that in the dataset, the proposed method has higher detection accuracy, the improved algorithm average accuracy improves 3.9%, the model size decreases 20%, effectively improves the detection effect of small objects, and realizes the lightweight of the network.

Enhanced 3D object detection for autonomous driving: A spatial-temporal alignment approach in Bird's Eye View scenarios

This paper presents a novel 3D object detection algorithm designed for Bird's Eye View (BEV) scenarios, which significantly improves detection capabilities by integrating spatial and temporal features. The core of our approach is the spatial-temporal alignment module that efficiently processes information across different time steps and spatial locations, enhancing the precision and robustness of object detection. We employ a temporal self-attention mechanism to capture the motion information of objects over time, allowing the model to correlate features across various time steps for identifying and tracking moving objects. Additionally, a spatial cross-attention mechanism is utilized to focus on spatial features within regions of interest, promoting interactions between features extracted from camera views and BEV queries. Our method also implements temporal feature integration and multi-scale feature fusion to enhance detection stability and accuracy for fast-moving objects and to capture multi-scale context information, respectively. The model employs an enriched feature set post alignment for 3D bounding box prediction, ascertaining the position, dimensions, and orientation of objects. We conducted experiments on two public datasets for autonomous driving nuScenes and Waymo Open Dataset, demonstrating that our method outperforms previous BEVFormer and other state-of-the-art methods in terms of detection accuracy and robustness. The paper concludes with potential future directions for optimizing the BEVFormer model's performance and exploring its application in broader scenarios and tasks.

Occlusion-guided multi-modal fusion for vehicle-infrastructure cooperative 3D object detection

In autonomous driving, leveraging sensor data (e.g. camera, LiDAR data) from both the vehicle and the infrastructure significantly improves perception capabilities. However, this integration traditionally results in increased demands on communication bandwidth. To address these challenges, we introduce Fusion2comm, an occlusion-guided feature fusion approach designed to optimize vehicle-infrastructure cooperative 3D object detection. Our innovative strategy employs an intelligent fusion of camera and LiDAR data to enhance the expressiveness of features. Subsequently, it leverages a segmentation model to extract foreground features and utilizes an occlusion-based selection of communication content, effectively easing bandwidth constraints. We propose a multimodal foreground feature fusion architecture that selectively processes and transmits critical information, substantially reducing irrelevant background data transfer. An innovative occlusion confidence-aware communication technique dynamically adjusts communication regions based on occlusion levels, ensuring efficient data exchange. Fusion2comm sets a new benchmark in the DAIR-V2X dataset, achieving an average precision of 71.25% with minimal bandwidth usage of 221.04 bytes. Our comprehensive experimental evaluations confirm that Fusion2comm substantially advances detection precision while simultaneously improving communication efficiency.

Radar-camera fusion for 3D object detection with aggregation transformer

Radar-camera fusion for 3D object detection with aggregation transformer

A Systematic Survey of Transformer-Based 3D Object Detection for Autonomous Driving: Methods, Challenges and Trends

A Systematic Survey of Transformer-Based 3D Object Detection for Autonomous Driving: Methods, Challenges and Trends

Diffusion Models-Based Purification for Common Corruptions on Robust 3D Object Detection.

LiDAR sensors have been shown to generate data with various common corruptions, which seriously affect their applications in 3D vision tasks, particularly object detection. At the same time, it has been demonstrated that traditional defense strategies, including adversarial training, are prone to suffering from gradient confusion during training. Moreover, they can only improve their robustness against specific types of data corruption. In this work, we propose LiDARPure, which leverages the powerful generation ability of diffusion models to purify corruption in the LiDAR scene data. By dividing the entire scene into voxels to facilitate the processes of diffusion and reverse diffusion, LiDARPure overcomes challenges induced from adversarial training, such as sparse point clouds in large-scale LiDAR data and gradient confusion. In addition, we utilize the latent geometric features of a scene as a condition to assist the generation of diffusion models. Detailed experiments show that LiDARPure can effectively purify 19 common types of LiDAR data corruption. Further evaluation results demonstrate that it can improve the average precision of 3D object detectors to an extent of 20% in the face of data corruption, much higher than existing defence strategies.

A Small-Object-Detection Algorithm Based on LiDAR Point-Cloud Clustering for Autonomous Vehicles.

3D object-detection based on LiDAR point clouds can help driverless vehicles detect obstacles. However, the existing point-cloud-based object-detection methods are generally ineffective in detecting small objects such as pedestrians and cyclists. Therefore, a small-object-detection algorithm based on clustering is proposed. Firstly, a new segmented ground-point clouds segmentation algorithm is proposed, which filters out the object point clouds according to the heuristic rules and realizes the ground segmentation by multi-region plane-fitting. Then, the small-object point cloud is clustered using an improved DBSCAN clustering algorithm. The K-means++ algorithm for pre-clustering is used, the neighborhood radius is adaptively adjusted according to the distance, and the core point search method of the original algorithm is improved. Finally, the detection of small objects is completed using the directional wraparound box model. After extensive experiments, it was shown that the precision and recall of our proposed ground-segmentation algorithm reached 91.86% and 92.70%, respectively, and the improved DBSCAN clustering algorithm improved the recall of pedestrians and cyclists by 15.89% and 9.50%, respectively. In addition, visualization experiments confirmed that our proposed small-object-detection algorithm based on the point-cloud clustering method can realize the accurate detection of small objects.

Voxel self-attention and center-point for 3D object detector

With the advancement of autonomous driving, the industrial demand for 3D object detection has continuously increased, leading to the development of anchor-based LiDAR object detectors reliant on convolutional neural networks (CNN). However, on the one hand, the poor receptive field of CNN limits the understanding of the scene. On the other hand, anchor-based methods cannot accurately predict the posture of objects in the steering. Therefore, in this paper, we propose the voxel self-attention and center-point (VSAC). Firstly, a voxel self-attention network is designed into VSAC to capture extensive voxel relationship. Secondly, considering the impact of feature weight on prediction results, the pseudo spatiotemporal feature pyramid net (PST-FPN) is proposed. Finally, we employ a center-point detection head to make the prediction direction closer to the real object during steering. The experimental results of VSAC on the widely used KITTI dataset, Waymo Open Dataset, and nuScenes dataset demonstrate its positive performance.

TS-BEV: BEV object detection algorithm based on temporal-spatial feature fusion

In order to accurately identify occluding targets and infer the motion state of objects, we propose a Bird’s-Eye View Object Detection Network based on Temporal-Spatial feature fusion (TS-BEV), which replaces the previous multi-frame sampling method by using the cyclic propagation mode of historical frame instance information. We design a new Temporal-Spatial feature fusion attention module, which fully integrates temporal information and spatial features, and improves the inference and training speed. In response to realize multi-frame feature fusion across multiple scales and views, we propose an efficient Temporal-Spatial deformable aggregation module, which performs feature sampling and weighted summation from multiple feature maps of historical frames and current frames, and makes full use of the parallel computing capabilities of GPUs and AI chips to further improve efficiency. Furthermore, in order to solve the lack of global inference in the context of temporal-spatial fusion BEV features and the inability of instance features distributed in different locations to fully interact, we further design the BEV self-attention mechanism module to perform global operation of features, enhance global inference ability and fully interact with instance features. We have carried out extensive experimental experiments on the challenging BEV object detection nuScenes dataset, quantitative results show that our method achieves excellent performance of 61.5% mAP and 68.5% NDS in camera-only 3D object detection tasks, and qualitative results show that TS-BEV can effectively solve the problem of 3D object detection in complex traffic background with lack of light at night, with good robustness and scalability.

Boundary points guided 3D object detection for point clouds

3D object detection in LiDAR point clouds is crucial for computer vision tasks such as autonomous driving. The two-stage approach with point cloud completion achieves remarkable performance by generating semantic surface points for foreground objects or learning bird’s eye view shape heat map labels. However, these methods require additional completion datasets, leading to substantial computation and memory demands. In this context, we propose a Boundary Points Guided 3D (BPG3D) object detection method that complements point cloud boundary information without the need for additional data. Specifically, we generate Region of Interest (RoI) boundary points to aggregate the neighbor voxel information at the RoI boundary during the refinement stage to complement the missing boundary information. Meanwhile, we design a Dual Feature Selection (DFS) module to adaptively fuse RoI grid point features and RoI boundary point features for bounding box refinement with negligible computational cost. Additionally, inspired by tensor decomposition theory, we use low-rank tensors to reconstruct high-rank tensors in the point cloud feature encoder to enhance contextual semantic information. The proposed method achieves 65.81% mAP on KITTI Test Set, obtaining a good trade-off between accuracy and efficiency.

DBSCAN and Yolov5 based 3D object detection and its adaptation to a mobile platform

This study presents a 3D object detection technology for mobile platforms and its application. Rather than an innovative high-performance model, we proposed a “useable” model for the robot industry at the current technology stage by combining various techniques. To reduce computation time, a 2D region proposal was obtained using a RGB image-based CNN model. By applying the DBSCAN clustering technique to the point cloud corresponding to the 2D region proposal, a method of obtaining a 3D region proposal was proposed. This allowed for 3D object detection using an RGB image dataset, which has been widely researched, while reducing the computation load to a level suitable for use in mobile robots. Furthermore, the 3D object detection was integrated into a ROS 2-based mobile platform, which was used to perform pedestrian-safe avoidance tasks and elevator button operation tasks. The performance was confirmed through experiments.

A streamlined framework for BEV-based 3D object detection with prior masking

In the field of autonomous driving, perception tasks based on Bird's-Eye-View (BEV) have attracted considerable research attention due to their numerous benefits. Despite recent advancements in performance, efficiency remains a challenge for real-world implementation. In this study, we propose an efficient and effective framework that constructs a spatio-temporal BEV feature from multi-camera inputs and leverages it for 3D object detection. Specifically, the success of our network is primarily attributed to the design of the lifting strategy and a tailored BEV encoder. The lifting strategy is tasked with the conversion of 2D features into 3D representations. In the absence of depth information in the images, we innovatively introduce a prior mask for the BEV feature, which can assess the significance of the feature along the camera ray at a low cost. Moreover, we design a lightweight BEV encoder, which significantly boosts the capacity of this physical-interpretation representation. In the encoder, we investigate the spatial relationships of the BEV feature and retain rich residual information from upstream. To further enhance performance, we establish a 2D object detection auxiliary head to delve into insights offered by 2D object detection and leverage the 4D information to explore the cues within the sequence. Benefiting from all these designs, our network can capture abundant semantic information from 3D scenes and strikes a balanced trade-off between efficiency and performance.

MS23D: A 3D object detection method using multi-scale semantic feature points to construct 3D feature layer

LiDAR point clouds can effectively depict the motion and posture of objects in three-dimensional space. Many studies accomplish the 3D object detection by voxelizing point clouds. However, in autonomous driving scenarios, the sparsity and hollowness of point clouds create some difficulties for voxel-based methods. The sparsity of point clouds makes it challenging to describe the geometric features of objects. The hollowness of point clouds poses difficulties for the aggregation of 3D features. We propose a two-stage 3D object detection framework, called MS23D. (1) We propose a method using voxel feature points from multi-branch to construct the 3D feature layer. Using voxel feature points from different branches, we construct a relatively compact 3D feature layer with rich semantic features. Additionally, we propose a distance-weighted sampling method, reducing the loss of foreground points caused by downsampling and allowing the 3D feature layer to retain more foreground points. (2) In response to the hollowness of point clouds, we predict the offsets between deep-level feature points and the object’s centroid, making them as close as possible to the object’s centroid. This enables the aggregation of these feature points with abundant semantic features. For feature points from shallow-level, we retain them on the object’s surface to describe the geometric features of the object. To validate our approach, we evaluated its effectiveness on both the KITTI and ONCE datasets.

Complete 3D Relationships Extraction Modality Alignment Network for 3D Dense Captioning.

3D dense captioning aims to semantically describe each object detected in a 3D scene, which plays a significant role in 3D scene understanding. Previous works lack a complete definition of 3D spatial relationships and the directly integrate visual and language modalities, thus ignoring the discrepancies between the two modalities. To address these issues, we propose a novel complete 3D relationship extraction modality alignment network, which consists of three steps: 3D object detection, complete 3D relationships extraction, and modality alignment caption. To comprehensively capture the 3D spatial relationship features, we define a complete set of 3D spatial relationships, including the local spatial relationship between objects and the global spatial relationship between each object and the entire scene. To this end, we propose a complete 3D relationships extraction module based on message passing and self-attention to mine multi-scale spatial relationship features and inspect the transformation to obtain features in different views. In addition, we propose the modality alignment caption module to fuse multi-scale relationship features and generate descriptions to bridge the semantic gap from the visual space to the language space with the prior information in the word embedding, and help generate improved descriptions for the 3D scene. Extensive experiments demonstrate that the proposed model outperforms the state-of-the-art methods on the ScanRefer and Nr3D datasets.

DopplerPTNet: Object Detection Network with Doppler Velocity Information for FMCW LiDAR Point Cloud

Abstract In the field of autonomous driving, LiDAR plays a crucial role in perception and detection. LiDAR based on Time-of-Flight (ToF) mode can only provide three-dimensional spatial coordinate information of point clouds. In point cloud object detection, the limited feature information of spatial coordinates to some extent restricts the further optimization and improvement of algorithm detection performance. However, LiDAR based on Frequency-Modulated Continuous-Wave (FMCW) mode can not only obtain the three-dimensional spatial coordinates of point clouds, but also directly measure the Doppler velocity information of points, effectively compensating for the limitation of relying solely on spatial coordinate information for object recognition. Therefore, based on the CARLA simulator, we construct the first FMCW LiDAR point cloud object detection simulation dataset, FMCWLidDet. What’s more, a novel 4D object detection algorithm, DopplerPTNet, is proposed based on the direct acquisition of point Doppler velocity information by FMCW LiDAR. The algorithm solves the problem of single spatial coordinate information feature in existing 3D object detection algorithms, which makes it difficult to further improve detection accuracy. The dataset is available at https://github.com/xilight123/FMCW-LiDAR-object-detection-dataset.

Explicit Incorporation of Spatial Autocorrelation in 3D Deep Learning for Geospatial Object Detection

Three-dimensional (3D) geospatial object detection has become essential for 3D geospatial studies driven by explosive growth in 3D data. It is extremely labor- and cost-intensive, though, as it often requires manual detection. Deep learning has been recently used to automate object detection within 3D context. Yet, addressing spatial dependency in 3D data and how it might inform deep learning for 3D geospatial object detection remains a significant challenge. Traditional models focus on the use of spatial properties, often overlooking color and contextual information. Exploiting these nonspatial attributes for 3D geospatial object detection thus becomes crucial. Our study pioneers explicit incorporation of spatial autocorrelation of color information into 3D deep learning for object detection. We introduce an innovative framework to estimate spatial autocorrelation, addressing challenges in unstructured 3D data sets. Our experiments suggest the effectiveness of incorporating spatial autocorrelation features in enhancing the accuracy of 3D deep learning models for geospatial object detection. We further investigate the uncertainty of such contextual information brought by diverse configurations, exemplified by the number of nearest neighbors. This study advances 3D geospatial object detection via using spatial autocorrelation to inform deep learning algorithms, strengthening the connection between GIScience and artificial intelligence and, thus, holding implications for diverse GeoAI applications.