245 
AUTOMATIC RECOGNITION OF RIVERS FROM LIDAR DATA BY PROFILE 
FACTOR 
Y. Lin a ' *, L. Yan a , Q. X. Tong a< b 
a Beijing Key Lab of SII&A, Institute of RS&GIS Research, Peking University, Beijing 100871, China, - 
lynnxi@yahoo.com.cn, lyan@pku.edu.cn, tqxi@263.net 
b Institute of Remote Sensing Applications, Chinese Academy of Sciences, Beijing 100101, China 
Commission VI, WG 1/2 
KEY WORDS: LiDAR, Automatic river recognition, Profile factor function, Edge extraction, Skeleton generation 
ABSTRACT: 
Laser infrared Detection and Ranging (LiDAR) has become one competitive remote sensing (RS) and photogrammetry technique. 
Extracting rivers’ distribution from LiDAR’s point cloud is one of its important research directions. But traditional methods are not 
completely suitable for the ranging data of LiDAR. Novel algorithm with profile factor as the kernel circle is proposed for rivers’ 
automatic recognition. Image unification, edge extraction and skeleton generation are the premier three steps. The profile factor of 
morphology can be expressed as shape functions for concrete judgement. Natural rivers’ profile is like “U” form, and artificial 
water-body’s can be simplified as “M” figure. While highway’s section can be considered as “W” shape. Then corresponding profile 
factor functions (PFF) can be established for determination. The experimental comparisons show that the results by the proposed 
algorithm are close to which from high-resolution RS images by manual interpretation. 
1. INTRODUCTION 
Laser infrared Detection and Ranging (LiDAR) has become one 
effective and competitive technique in remote sensing (RS) and 
photogrammetry fields, as LiDAR can acquire information with 
high ranging precision and high sampling density. The laser 
echoes stored in point cloud mode also supply the possibility of 
data processing as image. Among, extracting rivers’ distribution 
status from LiDAR data now is one important research point 
(David, 2006), as this function avails many applications of 
LiDAR, such as water and soil conservation, flood control, 
hydrology fine investigation, precision agriculture, pollution 
abatement and etc. 
For above applications, some river detection algorithms based 
on optical RS images have been developed (Lina, 2006). But 
the traditional methods are not completely suitable for LiDAR 
images which transformed from ranging data. LiDAR image 
pixels reflecting elevation feature is different with the spectral 
characteristics of RS images, i.e. SPOT5. Particular recognition 
algorithms need to be established for this new type of RS & 
photogrammetry mode. 
Simultaneously the requirement of non-human-interference and 
in-time process increases gradually, as more and more 
situations ask for immediate process and decision. The trends of 
on-satellite and on-board processing also advance the 
development of related concrete algorithms. Automatic process 
has become an ignorable factor in feature detection techniques, 
especially in flood control system. Novel algorithm for rivers’ 
recognition needs to be explored for LiDAR images. 
Based on the analyses above, this paper proposes one automatic 
recognition algorithm based on profile factor function (PFF) for 
rivers. The following contents include three parts. Firstly PFF is 
introduced. Secondly real data is assumed for testifying the 
algorithm. Finally conclusions are given. 
2. ALGORITHM 
The proposed river detection algorithm comprises three steps of 
pre-process, and category judgment by PFF is the last but 
kernel circle for rivers’ automatic recognition. 
2.1 Pre-process 
The pre-process is for constructing the frame suitable for PFF’s 
implementation. The first step is to acquire the corresponding 
images by equal-interval transition from ranging values to grey 
scales. Secondly edge extraction based on wavelet transform 
with the Prewitt operator is carried out. Then improved Hilditch 
thinning algorithm for rivers’ skeleton building is assumed. 
Actually between the edge extraction and skeleton generation 
false edges’ removal is one necessary circle. For RS images, 
this is especially difficult. In avoidance of struggling on the 
detailed and complex removal techniques’ discussion, such as 
distance threshold or circle removal (Guido, 2004), this paper 
describes this part without details’ introduction. The removal 
methods used in image processing of this paper will be 
introduced in authors’ another manuscript. 
2.1.1 Image unification 
The premise of all processing for river recognition is that point 
cloud has been regulated as clear grid set. Then the 
cumbersome point cloud can be converted into grey images for 
following operations. And this also avails the comparison of 
LiDAR images with high-resolution RS images. 
* Corresponding author.