To address the issue of low accuracy in extracting street tree information from point clouds, this paper proposes a multi-feature assisted MeanShift clustering ensemble method for dense street tree extraction. First, a stepwise classification approach, leveraging geometric and physical differences of target objects, is used to sequentially remove non-target point clouds—including ground, lawns, and shrubs—to eliminate interference. Then, an improved MeanShift clustering algorithm is developed using the kNN model. By incorporating multiple features such as reflectance intensity and spatial location, a fast, stable, and high-precision adaptive clustering ensemble method is constructed. Finally, a morphological estimation method is used to calculate geometric feature parameters of street trees, including tree height and crown dimensions. The proposed method is validated using dense street tree point cloud data acquired by the RIEGL miniVUX-SYS UAV LiDAR system in Yangshan Park, Nanjing. The experimental results show that the Individual tree height error is 0.025 m, the maximum crown radius error is 0.061 m, the projected area error is 1.321 m2, and the surface area error is 4.096 m2, outperforming traditional methods. The findings provide reliable support for the scientific management and maintenance of urban street trees, demonstrating promising application prospects.