Accurate detection of obstacles in 3D is an essential task for autonomous driving and intelligent transportation. In this work, we propose a general multimodal fusion framework FusionPainting to fuse the 2D RGB image point clouds at semantic level boosting object task. Especially, consists three main modules: multi-modal segmentation module, adaptive attention-based detector. First, information obtained Lidar based on approaches. Then results from different sensors are adaptively fused...

Multi-modal 3D object detectors are dedicated to exploring secure and reliable perception systems for autonomous driving (AD). Although achieving state-of-the-art (SOTA) performance on clean benchmark datasets, they tend overlook the complexity harsh conditions of real-world environments. With emergence visual foundation models (VFMs), opportunities challenges presented improving robustness generalization multi-modal detection in AD. Therefore, we propose RoboFusion, a robust framework that...

LIDAR and camera fusion techniques are promising for achieving 3D object detection in autonomous driving. Most multi-modal frameworks integrate semantic knowledge from 2D images into LiDAR point clouds to enhance accuracy. Nevertheless, the restricted resolution of feature maps impedes accurate re-projection often induces a pronounced boundary-blurring effect, which is primarily attributed erroneous segmentation. To address these limitations, we present <italic...

LiDAR-camera fusion can enhance the performance of 3D object detection by utilizing complementary information between depth-aware LiDAR points and semantically rich images. Existing voxel-based methods face significant challenges when fusing sparse voxel features with dense image in a one-to-one manner, resulting loss advantages images, including semantic continuity information, leading to sub-optimal performance, especially at long distances. In this paper, we present VoxelNextFusion,...

Multi-modal 3D object detectors are dedicated to exploring secure and reliable perception systems for autonomous driving (AD). However, while achieving state-of-the-art (SOTA) performance on clean benchmark datasets, they tend overlook the complexity harsh conditions of real-world environments. Meanwhile, with emergence visual foundation models (VFMs), opportunities challenges presented improving robustness generalization multi-modal detection in driving. Therefore, we propose RoboFusion, a...

In this technical report, we present our solution, named UniOCC, for the Vision-Centric 3D occupancy prediction track in nuScenes Open Dataset Challenge at CVPR 2023. Existing methods primarily focus on optimizing projected features volume space using labels. However, generation process of these labels is complex and expensive (relying semantic annotations), limited by voxel resolution, they cannot provide fine-grained spatial semantics. To address limitation, propose a novel Unifying...

Much recent effort has been devoted to employing Unmanned Aerial Vehicles (UAVs) implement airport-related tasks. However, a critical issue, collision avoidance, must be fully considered in this scenario. Herein, we study the efficient UAV navigation problem considering safety issue near an airport. In detail, first define safe separation between and airplanes according related aviation regulations. Thereafter, effective tree-based scheme for navigating proposed cope with extra uncertainties...

KNN is widely used in classification, but it could not gain good performance for multiattribute time series classifying. According to the characteristic of and shortage KNN, attributes weighted sample reducing classification approach-WRKNN proposed. Two major aspects are improved one give weight series; other reduce training set relative equal density based on distance. An equally distributed data obtained by approach, number samples decreased at same time, hence efficiency accuracy...

Self-supervised learning has made substantial strides in image processing, while visual pre-training for autonomous driving is still its infancy. Existing methods often focus on geometric scene information neglecting texture or treating both aspects separately, hindering comprehensive understanding. In this context, we are excited to introduce GaussianPretrain, a novel paradigm that achieves holistic understanding of the by uniformly integrating and representations. Conceptualizing 3D...

Accurate multi-view 3D object detection is essential for applications such as autonomous driving. Researchers have consistently aimed to leverage LiDAR's precise spatial information enhance camera-based detectors through methods like depth supervision and bird-eye-view (BEV) feature distillation. However, existing approaches often face challenges due the inherent differences between LiDAR camera data representations. In this paper, we introduce TiGDistill-BEV, a novel approach that...

Integrating LiDAR and camera information into Bird's-Eye-View (BEV) representation has emerged as a crucial aspect of 3D object detection in autonomous driving. However, existing methods are susceptible to the inaccurate calibration relationship between sensor. Such inaccuracies result errors depth estimation for branch, ultimately causing misalignment BEV features. In this work, we propose robust fusion framework called Graph BEV. Addressing caused by point cloud projection, introduce Local...

Interval data exist widely in industry and real life because the influence of uncertain, imprecise incomplete factors. In order to find crisp relationship among these data, nonlinear interval regression is used as an important avenue. former input output analysis, crude symmetrical estimation obtained for data. For asymmetrical set which error ranges differ upper lower ends, current approaches can't be depict exactly. this paper, analysis proposed first time. The two ends are studied...

Person re-identification has been extensively studied in recent years and made great progress. Many papers propose a lot of effective methods to improve the accuracy person re-identification. However, there are still many problems that remain unsolved. For example, persons often occluded by obstacles or other persons, leading loss complete information, changes behaviors postures make it difficult identify. In this paper, we algorithm repeatedly uses global feature information local for...

LiDAR and camera fusion techniques are promising for achieving 3D object detection in autonomous driving. Most multi-modal frameworks integrate semantic knowledge from 2D images into point clouds to enhance accuracy. Nevertheless, the restricted resolution of feature maps impedes accurate re-projection often induces a pronounced boundary-blurring effect, which is primarily attributed erroneous segmentation. To well handle this limitation, we propose general framework Multi-Sem Fusion (MSF)...

Accurate detection of obstacles in 3D is an essential task for autonomous driving and intelligent transportation. In this work, we propose a general multimodal fusion framework FusionPainting to fuse the 2D RGB image point clouds at semantic level boosting object task. Especially, consists three main modules: multi-modal segmentation module, adaptive attention-based detector. First, information obtained images Lidar based on approaches. Then results from different sensors are adaptively...

Most of the existing person re-identification methods usually follow a supervised learning framework and train models based on large number labeled pedestrian images. However, directly deploying these trained in real scenes will lead to poor performances, because target domain data may be completely different from training data, thus model parameters cannot well fitted. Furthermore, it is very time-consuming impractical label data. In order solve problems, we propose simple effective...