Extended Abtract
Introduction
 lidar data are raw point clouds that include three-dimensional coordinates and are taken by laser. These three-dimensional data can automatically produce the digital surface model (DSM), which plays an important role in earth-related applications. In practice, the processing of huge point clouds with the aim of modeling systematic errors, filtering, extracting complications, etc., requires a large amount of human interaction. In recent years, the use of this technology has made significant progress in revealing targets.
The ability of the laser pulse to penetrate through the leaves of the trees makes it possible to detect the targets located under the cover of the trees by LIDAR. Chang et al. (2010) for the first time presented a method for detecting objects located under trees. In this method, which operates based on the feature of multi-pulse LIDAR, statistical methods are used to reveal and extract the target points located under the trees. In the algorithm, the removal of the forest cover is done based on the morphological filter. The results of this algorithm show that the filtering process in this way has worked well in removing the tree cover and revealing the targets under the tree cover. The American Defense Advanced Study Agency (DARPA) in a project, called JIGSAW, has demonstrated the ability of lidar to identify targets located under trees.
Methodology
 In this study, a new method called band filtering algorithm is presented to use LIDAR point cloud data to detect buildings located under the cover of trees. This method includes two general steps; segmentation and disclosure of the targets. In the segmentation stage, first the cloud points are segmented into different segments based on height and distance parameters in windows with specific dimensions. Division is done in two directions with angles of zero and thirty degrees and is evaluated as an intermediate goal. Then, the ground and plant candidate points are separated from the building points, and in the next step, by comparing the height of the ground under the tree cover and its surroundings, the non-ground effects located under the tree cover are identified. Finally, using an innovative algorithm, edges are detected.
Experimental results
 In this study, the effectiveness of the proposed method was evaluated on lidar data related to an area of Santa Catarina state in Brazil. Due to the characteristic of receiving more than one pulse by the lidar sensor, the lidar pulses have the ability to penetrate under dense tree. This capability allows the sensor to detect targets located under tree. Considering the very high height accuracy of lidar data, it is possible to detect the targets located under the trees by comparing the height of the points under the tree cover and the points around the tree cover. In the proposed method in this article, first, the cloud points are segmented in different directions and appropriate labels are assigned to the segments. Then, by stacking the sections, the number of sections is optimized. Then it enters the edge detection stage in order to extract the targets located under the tree and if there is a target under the tree, by detecting the edges of the target, the desired target is extracted with the desired accuracy. to be Based on the experimental results of this study, the level under the receiver operating characteristic curve (ROC) of the proposed method is equal to 91%, which is 4% higher than the morphological filtering method. This evaluation shows the better performance of this method compared to the morphology filtering method.
In this study, in the available data, only buildings could be investigated as targets located under trees, and therefore, the threshold used in the proposed method is considered to be 2.7 to 3 meters, which the values are usually the height of the shortest buildings. By changing these threshold limits, it is possible to reveal other targets with different sizes. Therefore, as a suggestion, it is recommended to investigate and evaluate the proposed method for other possible purposes such as cars in other studies. It is also suggested to investigate and evaluate the detection of targets located under trees in sloping forest areas.