Comparison of Low-Cost 3D Reconstruction Techniques for Change Detection in Indoor Environments

This work evaluates low-cost 3D capture techniques for change detection in indoor environments. We compared a traditional Structure-from-Motion method (COLMAP) with a deep learning approach (CUT3R), using RGB images captured in two sessions of the same laboratory, one with visual changes and one without. The pipeline included 3D reconstruction, point cloud post-processing, alignment, and point-to-point (Cloud-to-Cloud) change detection. Results show that CUT3R produces visually richer reconstructions with significantly lower processing time but suffers from scale inconsistencies and object duplication, compromising precise change detection. COLMAP, although slower, maintains spatial coherence and provides more reliable identification of alterations. These findings indicate that traditional methods are still preferable for rigorous change detection, while deep learning approaches are advantageous for scenarios requiring fast or visually detailed reconstructions.