Pixel-level point cloud densification method for geometric clue perception in three-dimensional scene reconstruction

The invention relates to a pixel-level point cloud densification method for geometric clue perception in three-dimensional scene reconstruction, and aims to solve the problem of point cloud sparsenessand detail deficiency in a three-dimensional reconstruction process. By defining projection consistency logic, the correctness of a geometric relationship is measured and the attribution of pixels isadaptively adjusted to guide calculation of a spatial point cloud; pixel points input to an image set can be effectively integrated and a projection clue of a geometric space can be extracted, so that the density and accuracy of a three-dimensional reconstruction result can be effectively improved.