Neural radiation field (NeRF)-based novel view synthesis methods are gaining popularity for their ability to generate detailed and realistic images. However, most NeRF-based methods only use images to learn scene representations, ignoring the importance of depth information. The Zip-NeRF method has achieved impressive results in unbounded scenes by combining anti-aliasing techniques and mesh representations. However, the method requires a large number of input images and may perform poorly in complex scenes. Our method incorporates the advantages of Zip-NeRF and incorporates depth information to reduce the number of required images and solve the scale-free problem in borderless scenes. Experimental results show that our method effectively reduces the training time.And we can generate high-quality images and fine point cloud models using few images, even in complex scenes with numerous occlusions.

LiDAR point cloud is an important component of 3-D reconstruction, and high-precision 3-D point clouds provide accurate depth and scene structure. However, dynamic objects can form ghosting in the point cloud during data acquisition, causing data corruption. Most researchers have used the occlusion relationship between dynamic and static objects to achieve simple trajectory removal of dynamic objects in outdoor open scenes. However, in indoor scenes, due to the complex occlusion relationship, it is not possible to ensure the fine-grained, efficiency, and accuracy of the removal to achieve practical application levels. Here, we proposed an offline point cloud dynamic object removal method based on time visibility. We believe that dynamic objects can be regarded as a more obvious noise point, and it is feasible to distinguish dynamic and static objects from the time dimension under the premise of their unstable observation. Therefore, we proposed a multistage method for dynamic object removal. Specifically, we use hash-encoded voxels to express local space, encode local space’s time sequence occupancy count as a unified vector, and summarize comprehensive indicators of observation continuity and repeatability. This metric is used to evaluate the state of voxels, and then remove the dynamic part. In addition, we proposed a density-based binary classification method for more fine-grained removal tasks. Finally, we have validated the proposed algorithm’s robustness, advancement, and efficiency in multiple challenging indoor scenes.

Indoor high-precision maps are harder to obtain. Numerous researchers considered the extendibility of indoor plane features to constrain the poses and improve mapping consistency. However, the usage of parametric planar or line features requires complex modeling, and threshold-based matching methods are eager to lead to degradation. Indeed, planes can be more simply represented as line projections, with the projection direction perpendicular to the plane’s normal vector. In this article, we proposed a novel mapping optimization framework that incorporates building outline features specifically designed for multistory buildings. The framework consists of three main components: LiDAR bundle adjustment (LBA), global constraints based on building outline features, and factor graph optimization. The method extracts the ground only for horizontal correction, while the core idea is based on the assumption of vertical structure, projecting the point cloud contours of multistory buildings to the top view perspective for independent spatial pose correlation, which is a strong constraint and very effective. Finally, constraints are constructed based on keyframes and intermediate frames are refined by a factor graph. Through extensive experiments conducted on multistory buildings, we demonstrate that our algorithm significantly enhances mapping consistency and local accuracy compared to existing methods.

Neural radiation field (NeRF)-based novel view synthesis methods are gaining popularity. NeRF can generate more detailed and realistic images than traditional methods. Conventional NeRF reconstruction of a room scene requires at least several hundred images as input data and generates several spatial sampling points, placing a tremendous burden on the training and prediction process with respect to memory and computational time. To address these problems, we propose a prior-driven NeRF model that only accepts sparse views as input data and reduces a significant number of non-functional sampling points to improve training and prediction efficiency and achieve fast high-quality rendering. First, this study uses depth priors to guide sampling, and only a few sampling points near the controllable range of the depth prior are used as input data, which reduces the memory occupation and improves the efficiency of training and prediction. Second, this study encodes depth priors as distance weights into the model and guides the model to quickly fit the object surface. Finally, a novel approach combining the traditional mesh rendering method (TMRM) and the NeRF volume rendering method was used to further improve the rendering efficiency. Experimental results demonstrated that our method had significant advantages in the case of sparse input views (11 per room) and few sampling points (8 points per ray).

LiDAR–Inertial Odometry (LIO) has demonstrated robust real-time capability and efficient mapping performance compared to traditional terrestrial laser scanners. Although recent advances driven by public datasets have improved LIO stability under certain degraded conditions, existing studies still lack systematic validation in real-world scenarios. Key challenges include long-duration sequences, combined degradation factors, and seamless indoor-outdoor transitions. This gap limits their applicability to general-purpose, high-precision mapping. To bridge this gap, we present a large-scale, high-precision LIO dataset collected in four diverse real-world environments, covering areas ranging from 60 000 to 750 000 $\mathrm{{m}^{\mathrm{2}}}$. The dataset was collected with a custom backpack-mounted platform equipped with two multi-beam rotating LiDAR sensors, an industrial-grade IMU, and RTK-GNSS modules, achieving precise spatiotemporal alignment via hardware-level synchronization and rigorous calibration. The dataset encompasses a variety of environments, including structured buildings, tunnels, slopes, and urban scenes, to support seamless indoor-outdoor mapping. Each sequence has an average length of 1500 $\mathrm{m}$ and comprises over 15 000 frames per sequence, exceeding the scale of most public LiDAR–IMU datasets. We also propose a 6-DoF ground-truth generation pipeline that fuses SLAM-based optimization with RTK-GNSS anchoring. Using oblique photogrammetry integrated with RTK-GNSS measurements, we obtain georeferenced maps at a real-world scale. These maps are then used to align the LiDAR reconstructions and to validate the centimeter-level accuracy (<3 $\mathrm{c}\mathrm{m}$) of the generated trajectories. This dataset advances the state of the art in trajectory length, scene complexity, and ground-truth precision, providing a comprehensive dataset for evaluating LIO systems and improving their generalization to practical high-precision mapping scenarios.