Digital Signal Processing Reference
In-Depth Information
A Methodology for Ground Targets Detection
in Complex Scene Based on Airborne LiDAR
Jiafu Zhuang, Jie Ma
*
, Yadong Zhu, and Jinwen Tian
State Key Laboratory of Multi-spectral Information Processing Technology,
Huazhong University of Science and Technology, Wuhan, China
majie.hust@sohu.com
Abstract.
In this paper, an approach to detecting ground targets using LiDAR
point data is proposed. First, outliers are weeded out and point cloud is divided
into ground points and non-ground points. Second, the ground surface plane is
fitted by ground points and then the relative elevations of all non-ground points
are estimated. If the relative elevations of non-ground points exceed a prede-
fined threshold, they will be removed. Subsequently, a 3D region growing algo-
rithm based on the normal vector consistency is employed to generate potential
ground targets. Geometric information is used for further filtration of these po-
tential targets on the object level. Finally, the detection performance of the al-
gorithm is analyzed. The experimental results show that the method proposed is
effective.
Keywords:
LiDAR, Detection, 3D Region growing.
1
Introduction
An airborne LiDAR system usually consists of a platform and a scanning laser sensor
which are active sensors utilizing lasers to illuminate a scene and detectors to measure
the return signals. It can take the initiative, real-time 3D information to a wide range
of surface. The airborne LiDAR system will produce a large number of accurate 3D
coordinates of discrete points after the laser scanning missions. How to detect ground
targets from the large number of LiDAR points is mainly discussed in this article.
1
Although the point cloud gives a rich 3D information, many of the detection algo-
rithms grid point cloud into a height image, or directly use range image or intensity
image for segmenting and detecting the targets. HC Palm [1] extract edges, vertical
structure features using image processing methods for target detection. M. Himmels-
bach [3] divides point cloud into a number of grids. Each grid is assigned a value of
the maximum absolute elevation difference of points locating in that grid. The edge of
the target can then be extracted, around which indicates the existence of targets.
Tomas Chevalier [5] estimates the ground surface by the watershed algorithm and
then further estimate the relative height to ground of each non-ground points, follow-
ing the L-shapes fitting method to achieve the targets. Christina [2][6] uses a local
surface detection method to detect the targets.
* Corresponding author.
