To overcome the low efficiency and subjectivity of manual inspection in building facade geometric quality measurement and defect detection, this paper proposes an integrated method combining geometric measurement and unsupervised learning. First, a laser scanning binocular stereo vision system is developed to acquire high-precision three-dimensional (3D) point cloud data of building facades, and quantitative measurement of verticality and flatness is performed. Second, a dual-stream unsupervised anomaly detection model based on a self-supervised Vision Transformer (Self-Distillation with No Labels, DINO) and a Point Cloud Masked Autoencoder (Point-MAE) is constructed. Through multimodal feature fusion and a memory bank mechanism, facade defect detection and localization can be achieved without defect annotations. Finally, experiments are conducted on a self-constructed dataset. The experimental results show that the proposed method can accurately measure facade geometric quality and effectively detect surface defects, demonstrating good detection accuracy and engineering applicability.