An Omni-directional vSLAM based on Spherical Camera
Model and 3D Modeling
*
Guofeng Tong and Zizhang Wu NingLong Weng and Wenbo Hou
College of Information Science and Engineering College of Information Science and Engineering
Northeastern University Northeastern University
Shenyang, Liaoning Province, China Shenyang, Liaoning Province, China
tongguofeng@ise.neu.edu.cn 20060705wu@sina.com
*
This work is supported by NSFC Grant #61175031 and National High Technology Research and Development Program of China Grant #2009AA043902.
Abstract - This paper presents an efficient Omni-directional
Visual Simultaneous Localization and Mapping (vSLAM)
algorithm based on spherical camera model and 3D modeling. In
the paper, the robot has the ability of Omni-directional vision,
which makes the algorithm more adaptive in an unknown
environment. To get spherical panoramic images, we choose the
panoramic image acquisition and mosaic equipment (divergent
camera cluster). The improved SURF on spherical image, is
adopted for feature extraction and matching. According to the
theory of multiple view geometry of the spherical camera model,
the 3D modeling is conducted for the surrounding environment.
By using the feature points with high robustness, the location and
pose of the robot can be estimated. In the process of system
updating, the particle filter combined with Kalman filter is used
for it can perform well in a complex environment. The results of
numerical simulations and experiments have been included in
this paper to verify the performance of the proposed approach.
Index Terms – vSLAM. SURF. Spherical camera model.
Particle filter.
I. INTRODUCTION
Mobile robot simultaneous localization and mapping
(SLAM) is the core problem in the research of robots. Vision
based robot SLAM is an important method of the realization
of robot autonomous localization, perception of the
environment and scene understanding. In respect of the
integrity and richness of the information, vision based robot
SLAM has irreplaceable advantages, compared to the other
methods, such as laser, infrared, gyroscopes and so on.
However, the traditional vision based robot SLAM method,
has the disadvantage that the field of view is limited by using
the standard CCD camera to obtain information [1, 2]. So it
cannot make the robot get the overall environment information
at one time, which may lead to the missing of some visual
landmarks or other important information. This method will
lead to the robot response delay, and reduce the robot’s
adaptability for complicated environment.
If the panoramic camera is installed onto the autonomous
mobile robot, at the same time of the autonomous mobile
robot’s exploration unknown environment, the locomotive
robot can obtain holographic image and panoramic view
sports model [3]. This method can effectively realize the
exploration of the unknown environment and information
acquisition, shown as figure 1.
Fig. 1: The operation results of VR-SLAM
While the panoramic vision sensor, such as a fish-eye
camera, catadioptric cameras, PTZ camera, ladybug etc., has
the advantages of wide range of signal detection, fast and
complete information acquisition [4]. With only one camera, it
can observe the 360 degree panoramic environment, which is
centered by the robot. Not only this method is the fast access
to panoramic environment information, but can significantly
reduce the complexity of design system. Obviously,
panoramic vision based SLAM is an effective means of
obtaining the environmental information, understanding the
scene and building the environment map.
The fisheye lens method, image mosaic method, and fold
reflection imaging method based on the reflection principle of
curved mirror, are principal techniques to implement
panoramic vision. Using of ultra wide angle optical lens (can
be observed in the hemisphere which is closer to the object),
the fisheye lens method is complex and difficult to design.
And its demand of processing, assembling is high. The
distortion model does not meet the conditions of the
perspective projection, because the image distortion is serious.
Fold reflection imaging method, using discount reflection
optical element to realize panoramic imaging [5, 6], can well
solve the existing problems of the front two imaging scheme.
The method has become the hot research of issue. A
panoramic imaging system is composited of conventional
cameras and fold reflection optical element reflective surface.
The system makes use of the mirror reflecting the light in the
special light path to expand the view angle. The system is
simple in structure, easily installation, large view angle, high
speed. And the system has advantages of that image resolution
can achieve a satisfactory level according to the concrete
application.
Proceedings of the 10th
World Congress on Intelligent Control and Automation
Jul
6-8, 2012, Bei
in
, China
4551
978-1-4673-1398-8/12/$31.00 ©2012 IEEE