Single tree detection algorithm oriented to plantation forest inventories using mobile laser scanning data

Forest inventories are important in the management of forest resources and have long been conducted using traditional methods, which are generally expensive and time-consuming. With recent advances in terrestrial remote sensing technologies, including light detection and ranging (LiDAR) and photogrammetry, new paradigms have emerged to perform these tasks more quickly and accurately. However, some scientific advances are still required before these new alternatives can be implemented. Therefore, this study aimed primarily to develop a new algorithm capable of detecting individual tree trunks in plantations through point clouds captured by a mobile laser scanner (MLS), with high accuracy and low computational cost. The algorithm uses individual tree detection clustering methods with Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and random sample consensus (RANSAC) for circle fitting. Experiments were conducted using 13 point clouds, with 10 clouds of eucalyptus species plantings and 3 clouds of plantings of other species (mahogany, rubber tree and teak). The clouds were divided into three groups according to planting conditions, such as the presence of underbrush and tree alignment. All point clouds were processed on four computers using different configurations to evaluate the efficiency of the algorithm by measuring the execution time and resource usage. The algorithm achieved a trunk detection accuracy of 100% under ideal conditions and approximately 96% in more challenging scenarios.