A marked point process for automated tree detection from mobile laser scanning point cloud data

This paper presents a new algorithm for tree detection from airborne / mobile laser scanning or LiDAR point cloud data. The algorithm takes advantage of a marked point process to model the locations of trees and their geometries. The algorithm also uses the Bayesian paradigm to obtain a posterior distribution for the marked point process conditional on the LiDAR point cloud data. A Reversible Jump Markov Chain Monte Carlo (RJMCMC) algorithm is developed to simulate the posterior distribution. Finally, the maximum a posteriori (MAP) scheme is used to obtain optimal tree detection. This algorithm has been examined by a set of LiDAR point cloud data. The results demonstrate the efficiency of the proposed algorithm for automated detection of trees.

I. Introduction

Airborne/mobile laser scanner or light detection and ranging (LiDAR) sensor data have emerged in recent years as a leading source for automated extraction of various objects (e.g., buildings, trees, vehicles, terrain, etc.), particularly due to the direct measurements of surface topography both accurately and densely [1], [2]. To date, a variety of methods for tree detection and extraction have been proposed. For color infrared images, different tools have been developed. Some of them use pixel-based methods and give the delineation of the tree crowns, such as the valley following algorithms [3]. Other tools use an object-based method, by modeling a synthetic tree crown template to find the tree top positions [4], [5].