Segmentation of tree regions using data of a full-waveform laser

The new sensor technique for receiving full-waveform laser data delivers not only height information but also additional features like intensity. The echoes generated by reflecting objects are approximated by amplitude and width of a Gaussian kernel. These features are used for tree region segmentation. The inclusion of the available information in the environment around each 3d point of the cloud can be used to stabilize the segmentation process. Two kinds of environments spherical and cylindrical are discussed. The calculation of the covariance inside these environments is discussed without and with weighting all points by their measured intensity. An application of the described method is demonstrated for a typical scene. A method for the correction of the number of points inside an environment produced by the movements of the sensor platform is presented. Two proposals are described for the segmentation of trees. One of them uses the eigenvalues of the covariance matrix in an environment. The other method considers only the features of the echoes for each laser pulse without considering an environment.