COL 10(Suppl.), S11002(2012) CHINESE OPTICS LETTERS June 30, 2012
Ground target recognition based on imaging LADAR point
cloud data
Xiaofeng Li (ooo¡¡¡¸¸¸)
∗
, Jun Xu (MMM ), Jijun Luo (ÛÛÛÈÈÈ), Lijia Cao (ùùùáááZZZ),
and Shengxiu Zhang (ÜÜÜ???)
Xi’an Research Institute of High-tech, Xi’an 710025, China
∗
Corresponding author: xiaofeng li2006@126.com
Received August 15, 2011; accepted October 24, 2011; posted online April 18, 2012
We illustrate an approach to statistical model and sequential hypothesis designed to the automatic target
recognition (ATR) problem for active imaging LADAR. The key to this approach is using multihypothesis
sequential tests to reduce the number of target hypotheses under consideration as more observed data
are processed. The approach is potentially useful when sensor data are plentiful but computation time
and processing capability are constrained. We experimentally demonstrate that the proposed recognition
approach can resolve the military ground vehicle recognition problem of active imaging LADAR with a
high recognition rate.
OCIS codes: 100.3008, 280.3640.
doi: 10.3788/COL201210.S11002.
The battlefield scenario continues to grow in complexity
as the use of high-resolution sensors and precision strike
weapons has forced the incr eased use of concealment and
camouflage technology to improve weapon survivability.
Thus, automatic target recognition (ATR) capability is
becoming increasingly impo rtant to the Defense Commu-
nity. There are two fundamental challenges in ATR for
military applications . The first is the mass ive data loads
generated by modern high-resolution sens ors, which can
quickly escalate to unmanageable proportions. The sec-
ond challenge is camouflage, concealment, and deception,
this further increases the complexity of ATR and moti-
vates the need for robust techniques
[1]
.
A promising approa ch to tackle these challenges is
LADAR imagery
[2]
. Active imaging LADAR typically
provides both intensity a nd ra nge images. Intensity im-
ages give an indication of ta rget material. The r ange im-
ages provide explicit three-dimensional (3D) information
about targets. This 3D data contains geometric invari-
ant pro perties of targets and theory avoids the types of
distortions and ambiguities created by two-dimensional
(2D) sensor s. Additionally, active imaging LADAR pro-
vides foliage and camouflage penetration. This simplifies
foreground and background clutter removal, facilitates
target detection and segmentation, and yields higher
recognition rates and lower false alarms rates. There
is considerable interest in developing robust 3D ATR
technology a s LADAR sensors become more widely avail-
able, and many approaches have been developed fo r the
purp osed of target recognition based on LADAR data.
Recognition methods using LADAR intensity informa-
tion or the combined range and intensity information,
can be found for example in Refs. [3, 4], and here we fo-
cus on the methods only using LADAR range data which
usually represented as “point c loud”.
Previous works on automatic target recognition in
LADAR is to convert the 3D space to a mor e simple
2D coordinate space. These recognition algorithms take
advantage of the re latively large amount of processing
tools in the 2D image analysis field. Vasil et al.
[5]
used
spin images for 3D LADAR data re c ognition. In spin-
image-based representations every point in a cloud will be
associated with a representative 2D image that projects
surrounding points with a loca lly referenced coordinate
system. The efficiency of their approach is limited how-
ever, as it involves analyzing and classifying data through
several layers of filters and iterative processes in a point
by point basis. Gr¨onwall et al.
[6−8]
used general 3D
point scatterer in their work, which focuses on r ecog-
nizing ground vehicles. Their approach is base d on the
assumption that man-made objects of complex shape can
be decomposed into a set of recta ngles. Many other
LADAR range data ba sed object recognition approaches
are reviewed in Refs. [6, 9].
Recently, statistical approaches are usually found in
the literature dealing with ATR based on LADAR range
imagery
[10−14]
. Rather than treating data sets a s the out-
come of deterministic meas urement processes , statistical
approaches treat measurements as random variables un-
der particular distributions in order to model observation
uncertainty. The most important advantage of the sta-
tistical approaches is that once the prior knowledge and
the assumptions about the data are given, the statistical
approaches always give consistent and concise solutions
to problems of detection, classificatio n and parameter es-
timation by application of basic principles of statistical
inference.
One such approach revolves around the use of a min-
imum probability of error decision rule to compute the
likelihood that a se t o f 3D obse rvations (such as po int
cloud) arose from any one of a number of known pos-
sible targ e ts. Under this decision rule, however, likeli-
hoods must be calculated for every component point in
the cloud against each candidate target hypothesis. Ap-
plying this scheme to the practical application of r ecog-
nizing r e al-world vehicles, an individual point-wise like-
lihood ca lculation necessitates a surface integral over a
non-trivial figure. The number of these costly computa-
tions scale linearly with each data point processed and
each target hypothesis tested. Thus, it is very desirable
to reduce the number of times new likelihoods are evalu-
ated.
1671-7694/2012/S11002(4) S11002-1
c
2012 Chinese Optics Letters
下载后可阅读完整内容，剩余3页未读，立即下载
weixin_38501810
- 粉丝: 2
- 资源: 942
我的内容管理
展开
最新资源
- 深入理解23种设计模式
- 制作与调试：声控开关电路详解
- 腾讯2008年软件开发笔试题解析
- WebService开发指南：从入门到精通
- 栈数据结构实现的密码设置算法
- 提升逻辑与英语能力：揭秘IBM笔试核心词汇及题型
- SOPC技术探索：理论与实践
- 计算图中节点介数中心性的函数
- 电子元器件详解：电阻、电容、电感与传感器
- MIT经典：统计自然语言处理基础
- CMD命令大全详解与实用指南
- 数据结构复习重点：逻辑结构与存储结构
- ACM算法必读书籍推荐：权威指南与实战解析
- Ubuntu命令行与终端：从Shell到rxvt-unicode
- 深入理解VC_MFC编程：窗口、类、消息处理与绘图
- AT89S52单片机实现的温湿度智能检测与控制系统
资源上传下载、课程学习等过程中有任何疑问或建议，欢迎提出宝贵意见哦~我们会及时处理！
点击此处反馈
