Orthogonality-guided alignment of city-scale point clouds under strict memory budgets

Abstract

The rapid proliferation of three-dimensional scanning technologies has enabled the capture of city-scale point clouds with unprecedented detail and precision. These massive datasets are foundational for applications ranging from autonomous navigation and urban planning to the creation of comprehensive digital twins. However, the sheer volume of spatial data introduces severe computational bottlenecks, particularly concerning memory consumption during the alignment and registration phases. Traditional methods frequently exceed the memory capacities of standard computational hardware, necessitating severe downsampling that compromises structural fidelity. To address this critical challenge, this paper presents a novel methodology for the orthogonality-guided alignment of city-scale point clouds operating under strict memory budgets. By enforcing orthogonality constraints on the extracted feature spaces, the proposed framework minimizes redundant information and maximally preserves geometric distinctiveness without requiring massive memory overhead. The approach systematically decorrelates feature dimensions, ensuring that each parameter optimally contributes to the spatial alignment process. Comprehensive evaluations on simulated and real-world urban datasets demonstrate that the proposed orthogonality-guided technique achieves state-of-the-art registration accuracy while reducing memory utilization by substantial margins compared to contemporary alignment architectures. The results indicate that strict memory constraints do not preclude high-fidelity spatial registration when functional subspaces are optimized through rigorous geometric regularizations.