The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008
870
improved position of camera is then used as control point for
refine navigation data and then the improved position and
orientation for laser is calculated.
2.3 Positioning Accuracy from Our Experiments
Positioning accuracy is the integrated accuracy with navigation
sensors and mapping sensors. Mapping accuracy of some GCPs
is listed in comparison with terrestrial surveying, which is cm-
level accuracy. It is believed that our system mapping accuracy
can reach to about 30 cm order by our high-accuracy
positioning technology by fusion processing of GPS/IMU,
Stereo image sequence and laser point cloud. Our high-
accuracy positioning method also acquires good position
accuracy in the area of poor GPS signal and multi-path.
3. AUTOMATIC ROAD MAPPING BY FUSING IMAGE
AND LASER DATA
-#
The term automatic road mapping in this research is defined as
to automatic extract, recognize and position road objects from
collected data by vehicle-based mobile mapping system. The
main topic in this section is to discuss how to automatically
collect road spatial information by fusing stereo image and laser
data which are captured by our vehicle-borne multiple sensor
mobile mapping system. The keywords of this section are data
fusion and automatic mapping; data fusion is processed for
better automatic mapping, and automatic mapping is done for
more efficient and less costly generating high-definition road
spatial data.
3.1 Why need fusion of image and laser data
For automatic extraction and recognition, typically, there are
image-based approach, laser-based approach, and fusion
approach. Image-based approach base on digital image
processing thesis, utilize color, shape, texture to recognize
object. Image-based approach has very long development
history; many of the techniques were developed in the 1960s at
the Jet Propulsion Laboratory, MIT, Bell Labs, University of
Maryland, and a few other places. Although well developed
image-based approach has many successful applications, it has
some inherent drawbacks as below:
1. Occlusion problem own to it’s center perspective projection;
2. Be sensitive to environment conditions
such as light and shade;
3. Mosaic from piece of images is costly (FIgure3-a);
4. Stereo matching based 3D positioning is difficult
for full automation.
Laser scanner is active sensor, so that laser-based approach is
insensitive to environment conditions. Because laser scan object
through point by point, exclusion problem of laser-based
approach is not so serious with comparison of image-based
approach. And laser scanner can acquire the continuous 3D
point cloud, so there are no mosaic problem and also no stereo
matching problem. From the above analysis, it is obvious to say
that laser-based approach just can solve those unsolvable
problems by image-based approach. However, laser-based
approach also has its inherent drawbacks such as no gray or
color information so that recognition becomes difficult. BUT
that is just strong point of image-based approach.
The Figure 3 shows some typical merits of road object
automatic extraction by fusion method, (b) demonstrates that
color-rendered laser point cloud is easy to extraction road data
without mosaic processing; (c) shows that color-rendered laser
point data can be used for detecting road mark; (d)(e) show
laser point cloud can support road object detection and
recognition using its robust shape due to active sensor.
Figure3. Comparison of image and laser data for automatic
object recognition and extraction
3.2 Automatic Road Object Extraction by Fusing Stereo
Image Sequence and Laser Point Cloud
From the above discussion, it is easy to know data fusion
method can get over those inherent drawbacks of individual
approach. Our developed approach is based on fusion of image
and laser approach. The data origin is laser range data, stereo
image sequences, and automatic acquiring of road spatial data is
done by fusion method. The processing flowchart of fusion
method is drawn in figure 4. The processing includes five key
steps as 1.Direct geo-referencing; 2.Laser point rendering;
3.Image Positioning; 4. Laser point cloud based candidate
extraction using shape, color and position and 5. Image and
laser data fusion based final robust extraction.
Direct Geo-referencing
Direct geo-referencing is to identify the spatial position of the
objects scanned by mapping sensors at any time while the
vehicle is moving with reference to a common coordinate
system. The detailed information can be reviewed in chapter 5.
By high-accuracy positioning technology, we have calculated
high-accuracy instant sensor posture (X,Y,Z, (p,co,ic) when laser
point or image is recorded. Based on the posture, laser 3D point