- Remote sensing forest inventory gained increased attention in the last decade triggered by the decrease in the price of sensors and the explosion of data availability and formats. Evermore, the constant advances in the hardware processing the data emphasized the necessity to develop algorithms to extract forest relevant information from the images or point clouds acquired using remote sensing techniques. The present research develops a method to identify individual trees and estimate stem dimensions from a terrestrial point cloud by integrating three procedures: semantic segmentation, density-based clustering, and sector-based attribute measurement. The semantic segmentation, implemented by modifying the PointNet++ algorithm, classifies the points as ground, stem and crown, whereas the density-based clustering, implemented by adjusting the DBSCAN algorithm, delineated individual trees from the point cloud. To test the algorithm’s robustness, universality and accuracy I have used two lidar datasets: the multiple scan international benchmarking and the McDonald-Dunn Research forest. For the international benchmark data, the dominant and codominant trees were more than 95% correctly identified, whereas the completeness was at least 94%, except for one of the plots which was 84%. Most of the missing trees are caused by boundary effect and inconsistent point density, but the proposed method still outperformed the state-of-the-art tree measuring algorithms using point clouds. The bias in estimation of the diameter at breast height (DBH) is < 3 mm (similar to the best results of 1 mm), except in one plot where the bias reached 6.2 mm. However, the variability of the proposed method is almost half (i.e., <14 mm) compared with the best result (i.e., 20 mm). The situation was mirrored for height estimation, with bias <7% (except for one plot which exhibited 11% bias). The application of the algorithm to the McDonald-Dunn Research Forest supplied a completeness of 94.25% and a 96.47% correctness, even though the algorithm was developed on completely different species and the trees were not seen from all sides (i.e., incomplete description of the stem). The diameter bias at 1.5 m and 2.5 m was at most 1.3%, confirming the accuracy and generality of the algorithm. The algorithm provides operational worthy results in tree recognition and diameter estimation for point clouds that almost completely describe the trees. However, the algorithm seems to be sensitive to the stem coverage with points, which should be addressed in future research.