A method for 3D reconstruction of ancient buildings driven by crowdsourced images

Ancient buildings are in urgent need of protection owing to erosion by multiple hazards, such as structural aging and surface corrosion. Traditional 3D reconstruction faces issues such as high cost, limited coverage, geometric distortion, and texture degradation caused by the multi-scale quality heterogeneity of crowdsourced images. This study proposes a three-stage crowdsourced image optimization framework that integrates multimodal features. First, an adaptive cascaded quality screening architecture was constructed to achieve data pre-screening; second, an XGBoost dual-dimensional image feature classifier was proposed; finally, a multi-domain collaborative quality assessment model was established. Taking the Yingxian Wooden Pagoda as an example, experiments show that this method increases the surface density by 25.7%, improves texture clarity by 29.4%, and reduces SOR (Statistical Outlier Removal) by 14.7%. It effectively solves the problem of geometric-texture coupled distortion and provides a novel, efficient, and practical technical approach for accurate digital protection of complex ancient buildings.