Extraction of individual buildings is a key step in 3D reconstruction. hrough the traditional excess green index method to classify vegetation points and then extract individual units results in many building points being incorrectly classified as vegetation points, making it difficult to ensure the integrity of the extracted individual buildings. In response to this issue, an improved algorithm for the extraction of individual buildings was proposed. Firstly, the vegetation points were classified by combining the point cloud green index and normal vector information. Then, considering the difference in spatial density distribution between the misdivided building points and the vegetation points, spatial density detection was used to identify and reclassify the misequinox points as non-vegetation points. Finally, density clustering was used to segment individual buildings, resulting in more complete individual building structures. Experiments with two sets of data having significant differences in vegetation distribution show that the improved method proposed in this paper can extract individual buildings more completely, effectively avoiding excessive misclassification by the excess green index method. The method demonstrates significantly better integrity and quality accuracy than the Excess Green Index method, with minimal fluctuations in correctness.