内窥镜三维重建中的多尺度注意力机制特征匹配OA

3D reconstruction serves as a key foundation for precise minimally invasive surgery.However,poor texture and insufficient lighting in cavities pose significant challenges to feature matching,a crucial part of 3D reconstruction.a hierarchical structure-enhanced feature matching method is proposed.A C-SuperPoint feature extraction model,integrated with multi-scale attention mechanisms,enhances detection in weak texture areas,increasing the number of extracted feature points by 33.70%on average.Additionally,a P-LightGlue feature matching algorithm based on the transformer architecture performs initial rough matching and refines results through progressive consistency sampling,effectively filtering mismatches and fitting the model.This approach achieves up to 98.72%accuracy and a minimum average matching error of 1.065 1 pixels across multiple datasets.The results confirm that C-SuperPoint can effectively improve feature extraction performance,while P-LightGlue significantly enhances the accuracy and reliability of image processing in minimally invasive surgery.