Search

Data filtering and image segmentation are two crucial steps in individual tree information extraction based on airborne LiDAR data. In this paper, a new image segmentation method was proposed to isolate individual tree and extract tree height and size information of the tree crowns, which is the fusion of marker-controlled watershed segmentation and valley-following segmentation. This method can perform edge detection effectively compared to methods before because a special detection constraint was added into it. A real data test was given to testify its validation.

The objective in this study is to obtain the accurate tree crown model with complex structure from airborne lidar data for latter feature extraction. The segmentation of tree crown was implemented in several phases. First, the relatively high vegetation points were filtered out from the original tree dimensional point cloud by lidar data processing software. These vegetation points were interpolated in a grid, and then lowpass filter and highpass filter method were utilized to smooth the noise and sharp the crown edge respectively. In the next phase, the points were transformed to be a grayscale image, and the contrast of the image was enhanced by a contrast stretch algorithm to help the segmentation in latter step. Before the watershed segmentation was used to segment the tree crowns, the opening and closing operation in morphology were operated on the image to optimize the segmentation. Finally, a satisfying segmentation result was shown compared to the result which the contrast stretch algorithm wasn't operated on the image, even the overlapped tree crowns were segmented successfully in our test.

In this article, we use the height difference between the first pulse and last pulse recorded by the airborne laser scanning system to implement a strategy for detection of trees, especially for relatively tall trees. The airborne lidar data we use include 3D coordinates and intensity information of first pulse and last pulse respectively, and the image file of the test area are also available. ArcGIS version 9.0, has been used as the implementation platform as which as effective data processing function and outstanding performance of data analysis such as TIN interpolation and 3D display. To extract trees from the raw data, the data is loaded into ArcGIS to create shapefile recognized by ArcGIS at first, and the corresponding attributes, coordinates, intensity and height difference computed by programming are stored. Then we select the data accord with the filtering condition, a threshold for height difference is given. All the comparisons of 3D coordinates and intensity occur between point data in the same row, in other words, between the first pulse and last pulse. We accept the first pulse which satisfies the height difference threshold as vegetation points. At last, a considerable result of detection of trees has been achieved.