Estimation of number of trees using computer vision from UAV images

Unmanned Aerial Vehicles (UAV) are used in many fields such as military, agriculture, security, monitoring, emergency aid and tourism. In this thesis, the number of trees of some regions in Siirt University Kezer Campus area is estimated from high resolution colored RGB (Red, Green, Blue) images obtained from UAV. After the orthophoto map of the images taken from the air with the UAV in the designated areas within the campus area, the region of interest of each object was determined in HSV (Hue Saturation Value), RGB (Red Green Blue) and GRAY (Gray) formats of these orthophoto maps. A data set was created containing learning, validation and test datasets from regions of interest. This data set was tested by using various machine learning algorithms. In the last stage, the real numbers of the trees were obtained by manual method and a reference model was formed and the accuracy analysis was made by comparing the numbers of the trees obtained with the automatic method and the reference data. The developed software can be used in future projects for calculating the number of other objects such as vehicles, people and buildings.