Algorithms for the Automatic Detection of Trees in Laser Scanner Data

Laser scanner data can be used to describe the state of forests and its changes as a base for forest management activities. An automatic evaluation of typical forest parameters using laser scanner data enables the detection of trees with minimal user interaction. This paper describes an algorithm for detecting trees in a semi-automatic way. The detection runs in several steps beginning with filtering the scanned data to eliminate outliers in the point cloud. In the second step a digital terrain model (DTM) is generated. Based on the DTM, horizontal layers with a constant distance above the terrain are generated. Besides the interesting tree stems, these layers contain other objects such as bushes or tree crowns. To filter such objects in the layers, several methods are used. The residual tree stems in the layers are mapped as circle rings, which are detected by using a Hough-transformation and measured accurately by using a fit circle algorithm and a fit ellipse algorithm. For these standard pattern algorithms the layers are mapped on a regular raster and are saved in an image format with multi channels according to the different layers. The resulting set of circle parameters and the set of ellipse parameters contain the dimension and the position of the tree rings in a mixed-up form which is not sorted by individual trees. To rank them, the intersection of the tree ring area is used. In a final step a validation check of the resulting trees is done with a linear regression. For the use of this method in a standard application the processing time is essential. The listed time span for the process sequence show that the filter methods for the raw point cloud and the extraction of layers out of the filtered point clouds are too slow. To improve running time for these methods another data concept for holding the raw point cloud is essential and has to be developed. Furthermore, the data volume for raw data and results is stated.