LIDAR and camera fusion techniques are promising for achieving 3-D object detection in autonomous driving (AD). Most multimodal 3-D object detection frameworks integrate semantic knowledge from 2-D images into 3-D LiDAR point clouds to enhance detection accuracy. Nevertheless, the restricted resolution of 2-D feature maps impedes accurate reprojection and often induces a pronounced boundary-blurring effect, which is primarily attributed to erroneous semantic segmentation. To address these limitations, we present the multi-sem fusion (MSF) framework, a versatile multimodal fusion approach that employs 2-D/3-D semantic segmentation methods to generate parsing results for both modalities. Subsequently, the 2-D semantic information undergoes reprojection into 3-D point clouds utilizing calibration parameters. To tackle misalignment challenges between the 2-D and 3-D parsing results, we introduce an adaptive attention-based fusion (AAF) module to fuse them by learning an adaptive fusion score. Then, the point cloud with the fused semantic label is sent to the following 3-D object detectors. Furthermore, we propose a deep feature fusion (DFF) module to aggregate deep features at different levels to boost the final detection performance. The effectiveness of the framework has been verified on two public large-scale 3-D object detection benchmarks by comparing them with different baselines. And the experimental results show that the proposed fusion strategies can significantly improve the detection performance compared to the methods using only point clouds and the methods using only 2-D semantic information. Moreover, our approach seamlessly integrates as a plug-in within any detection framework.

Depth completion of forward objects is critical to safe and effective autonomous driving as it can estimate dense-depth from sparse LiDAR (i.e., Light Detection and Ranging) data and RGB images. The study proposes a self-supervised depth completion method to address the deficiency in depth sensing images caused by the uneven distribution of point-clouds generated by long-range LiDAR. The proposed method uses a stacked-hourglass-network structure as the backbone, which allows to perform multi-level deep fusion to extract features from uneven point-clouds. In addition, a fusion module is introduced to fuse global-local features from different stages of the network output, and a pooling guidance mechanism is designed to guide network convergence through adaptive pooling guidance depth construction. Field experiments were conducted on the Tsuen Wan Line in Hong Kong to verify the performance of the proposed network. The results indicate that the proposed network outperforms the benchmark self-supervised depth completion methods, with a root mean squared error (RMSE) of 8165.673mm and a mean absolute error (MAE) of 2596.418mm. Specifically, the RMSE and MAE were improved by 4.7% and 2.4%, respectively, compared to the benchmark methods.

Navigating complex urban intersections remains a significant challenge for autonomous vehicles due to highly dynamic and dense traffic environments. This paper presents a novel trajectory planning framework designed to address these challenges by integrating a dual-layer probabilistic rad-lane intention prediction model with an efficient sampling-based trajectory planner. The proposed dual-layer model comprises three components: short-term vehicle kinematic prediction-based road-level intention inference, Interactive Multiple Model (IMM)-based lane-level intention inference, and target-lane-based trajectory generation. This architecture broadens the application scope and validates the effectiveness of the IMM-based multi-model probabilistic fusion framework in urban intersection scenarios. By incorporating both road-level and lane-level contextual information, the proposed trajectory prediction method significantly enhances prediction accuracy. Additionally, a longitudinal sampling strategy based on predefined maneuver modes is employed to improve the probabilistic completeness of existing parametric curve-based sampling techniques, facilitating rapid and effective obstacle avoidance. A notable advantage of the proposed sampling strategy is its ability to efficiently and probabilistically generate safe and feasible trajectories in challenging intersection scenarios. Extensive simulation results demonstrate that the proposed framework outperforms existing methods in terms of both efficiency and safety in urban intersection vehicle conflict scenarios.

Lunar exploration is progressing rapidly, with craters serving as critical landmarks for precise spacecraft navigation and landing. Although deep learning has enhanced conventional crater identification techniques, persistent limitations include significant computational overhead and pronounced reliance on high-quality data. This study introduces SparseDLE, an efficient real-time crater detector built upon an optimized DETR framework. The method introduces a sparse query selection strategy to progressively refine results from coarse to fine; proposes a morphable grid attention moduleto enhance algorithmic efficiency; and designs a quality match optimizer to improve model accuracy. Experiments demonstrate that the proposed algorithm achieves an F1-score of 84.49%, outperforming existing state-of-the-art methods.

Planetary photogrammetry remains an important technical means of producing high-precision planetary maps. High-quality control networks are fundamental to successful bundle adjustment. However, current software tools used by the planetary mapping community to construct and edit control networks exhibit very low efficiency. Moreover, redundant and invalid control points in the control network can further increase the time required for the bundle adjustment process. Due to a lack of targeted algorithm optimization, existing software tools and methods are unable to meet the photogrammetric processing requirements of massive planetary remote sensing images. To address these issues, we first proposed an efficient control network construction framework based on approximate orthoimage matching and hash quick search. Next, to effectively reduce the redundant control points in the control network and decrease the computation time required for bundle adjustment, we then proposed a control network-thinning algorithm based on a K-D tree fast search. Finally, we developed an automatic detection method based on ray tracing for identifying invalid control points in the control network. To validate the proposed methods, we conducted photogrammetric processing experiments using both the Lunar Reconnaissance Orbiter (LRO) narrow-angle camera (NAC) images and the Origins Spectral Interpretation Resource Identification Security Regolith Explorer (OSIRIS-REx) PolyCam images; we then compared the results with those derived from the famous open-source planetary photogrammetric software, the United States Geological Survey (USGS) Integrated Software for Imagers and Spectrometers (ISIS) version 8.0.0. The experimental results demonstrate that the proposed methods significantly improve the efficiency and quality of constructing control networks for large-scale planetary images. For thousands of planetary images, we were able to speed up the generation and editing of the control network by more than two orders of magnitude.