Low-altitude UAV photogrammetry is increasingly used for 3D reconstruction of road environments. However, sparse point clouds produced by Structure-from-Motion (SfM) often suffer from occlusions and feature matching failures, leading to incomplete surfaces. We propose Direction-Aware Void Completion (DAVC), a geometric method that improves point cloud continuity by leveraging local directional cues. DAVC identifies deficient regions, extracts dominant directions via Principal Component Analysis (PCA), and extrapolates new points along structural lines, followed by density-aware filtering to avoid redundancy. Experiments on real-world UAV datasets show that DAVC achieves a mean completion error of 3.8 cm, outperforming Moving Least Squares, Point Completion Network, and GRNet. It improves surface recovery by 39%, reduces discontinuities by 34%, and runs 2.1 × faster than deep learning-based methods. Without requiring image data or model training, DAVC offers an efficient and structure-consistent solution for large-scale UAV-based road reconstruction.