Trees in green areas offer numerous benefits for the environment and human health. Up-to-date, information about trees in urban green areas is crucial for sustainable urban planning. Traditionally, the detection and inventory of urban trees have been conducted through field surveys and terrestrial measurements. However, this labour-intensive approach can be replaced with the more efficient LiDAR (Light Detection and Ranging) systems, an active remote sensing technology. Urban trees can be quickly and automatically determined using 3-dimensional (3D) LiDAR point cloud data. The objective of this study is to acquire trees in densely populated areas of large cities using raw LiDAR data. The urban study area was chosen in the Fatih district of Istanbul, which includes Sultanahmet Square, a site registered on the UNESCO World Heritage List. To detect urban trees, initially, eight classes representing the ground surface were obtained from LiDAR data with a point-based classification approach which is called hierarchical rule-based classification, and the high vegetation class was separated from the other classes. Noise points, which did not correspond to urban trees within the high vegetation class, were removed using the machine learning-based Density-Based Spatial Clustering of Applications with Noise (DBSCAN) clustering algorithm. The remaining high vegetation points were subsequently segmented using the machine learning-based Mean Shift clustering algorithm to obtain individual tree crowns. An accuracy assessment was conducted through completeness and correctness analyses, demonstrating the effectiveness of the proposed point-based approach for the automatic detection of urban trees from LiDAR data. According to the proposed Mean Shift clustering approach, the completeness was 60% and the correctness was 77.42% in test area A, while in test area B, the completeness was 62.30% and the correctness was 80.85%. The much higher completeness (78.26%) and correctness (100%) values were obtained for street trees with regular structure in test area B in comparison with the proposed Mean Shift clustering approaches.