Objective Light Detection and Ranging (LiDAR) technology is widely used in three-dimensional scene reconstruction and perception due to its ability to acquire high-precision geometric measurements. However, multi-view point cloud data collected in large or complex environments usually suffer from several spatial limitations. Point cloud frames captured from different viewing angles have independent coordinate systems. They also contain severe data occlusions, non-uniform point densities, and limited visual information. These issues degrade reconstruction quality and constrain practical engineering tasks such as target identification, geometric analysis, and structural monitoring. To resolve these challenges, this paper presents an integrated framework for multi-view point cloud registration and colorized visualization designed for large-scale operational environments. The principal goal is to eliminate alignment errors across different sensor poses while producing clear, continuous, and high-contrast color visualization models across multiple observation scales.

Methods  The proposed framework contains two main processing stages: a coarse-to-fine registration module and an orientation-guided color mapping scheme.

To unify multi-view point clouds into a single reference frame, a two-step registration strategy is established. In the candidate searching phase, rather than evaluating spatial Euclidean distances alone, a corresponding point searching mechanism is constructed by combining local geometric feature descriptors with laser reflectivity intensity values. For coarse registration, Singular Value Decomposition (SVD) is performed on the matched point pairs to calculate the initial rigid transformation matrix containing rotation parameters R and translation parameterst. This operation aligns separate point cloud frames into a shared spatial frame without requiring precise initial position estimations. Following coarse alignment, a fine registration step is conducted using the Iterative Closest Point (ICP) algorithm with point-to-plane distance constraints. The ICP procedure continuously minimizes the distance between corresponding points and target local surface planes, updating the transformation parameters iteratively until the spatial error converges below a preset threshold.

To improve visualization quality and remove visual artifacts caused by lighting variations or surface material shifts, a new color rendering method based on local surface dip direction is introduced. For every point in the registered dataset, the local surface normal vector \left(\mathit\mathit\mathrm\mathitn_\mathrm\mathitx,\mathit\mathrm\mathitn_\mathrm\mathity,\mathrm\mathitn_\mathrm\mathitz\right) is estimated by calculating the eigenvalues and eigenvectors of the spatial covariance matrix within its k-nearest spatial neighbors. Based on the components of the surface normal, the local dip angle and dip direction are calculated. The dip angle measures the inclination of the local surface relative to the horizontal plane, while the dip direction indicates the azimuth angle of the maximum slope. To generate a continuous visual representation, the 360-degree continuous values of the dip direction are mapped into the periodic Hue channel of the Hue, Saturation, Value (HSV) color space. Because the Hue component is inherently circular, this transformation forms a smooth color transition across continuous topological surfaces. It effectively prevents sharp color discontinuities caused by material changes, view point changes, or ambient light shifts during multi-view data fusion.

Results and Discussions The performance of the proposed registration and colorization system was evaluated through two field experiments representing different operational scales and acquisition configurations.

The first experiment evaluated a single station multi view scanning setup covering a commercial vehicle. This short range scenario focused on high-curvature metallic surfaces, dense reflection variations, and short distance occlusions. The second experiment evaluated a multi station single view scanning scenario targeting a rocket launch facility. This large scale scenario involved broad physical space, long measurement distances, structural shadows, and structural occlusions.

Experimental results show that combining geometric positions with intensity data during corresponding point selection improves matching stability, preventing convergence to local minima during SVD initialization. The subsequent fine ICP optimization reduces the global root mean square error (RMSE) across point correspondences down to sub-centimeter accuracy, eliminating registration ghosts and structural gaps in both the vehicle and launch pad models. In visual evaluations, conventional intensity color mapping generated bright spots, shadows, and inconsistent colors along view seams. In contrast, the dip direction color mapping highlighted subtle structural angles, structural edges, and slanted surfaces on both the vehicle body and launch structures. The structural elements remained distinct regardless of viewing distances or spatial scale changes.

Conclusions The experimental evaluation on vehicle scanning data and rocket launch site datasets confirms that the proposed approach successfully achieves high-precision point cloud alignment and complete geometric reconstruction. By integrating SVD and ICP algorithms with dip direction HSV color rendering, the system overcomes coordinate inconsistencies, spatial occlusions, and visual artifacts present in traditional multi-view point cloud fusion. The rendered three-dimensional models offer continuous, clear, and easily identifiable geometric features across different scales. Therefore, this paper provides a robust and reliable technique for multi-view three-dimensional reconstruction, target recognition, and structural assessment in complex real-world environments.