Accurate segmentation of individual trees from Airborne Laser Scanning (ALS) point clouds is essential for urban greening management, ecological conservation, and biodiversity assessment. However, the complex canopy structures of subtropical broadleaf forests, often characterized by multiple peaks, pose significant challenges for existing segmentation algorithms, leading to prevalent over-segmentation. To address this issue, we propose a novel tree segmentation framework that integrates high-resolution RGB imagery with airborne LiDAR point clouds, enhancing the extraction of individual trees in subtropical broadleaf forests. Our method first employs high-resolution imagery to delineate canopy boundaries, which serve as constraints to refine the clustering of supervoxel-segmented point clouds. Furthermore, to mitigate both over- and under-segmentation, an optimization step is introduced based on geometric shape features of tree crowns. Experimental validation conducted in Shenzhen, China, demonstrates the effectiveness of our approach, achieving an average recall of 0.902, precision of 0.890, and F1-score of 0.906 across two study areas. Compared to conventional tree segmentation techniques, our method improves recall, precision, and F1-score by 9.6%, 12.2%, and 13.3%, respectively. These results highlight the advantages of integrating multi-modal remote sensing data for fine-grained tree segmentation in complex forest environments.