Transportatio 
n Object 
U 
Imaging and 
GPS/INS 
Georeferencing 
4 
Figure 4. Verification and updating of transportation objects 
Since the images are all georeferenced, the back projection of an 
object from a map database onto images becomes straight forward. 
The transportation objects to be examined are firstly queried from an 
existing map database. The position of a desired object is projected 
onto the corresponding images using photogrammetric collinear 
equations. A global coordinate system, i.e., UTM coordinate system, 
is used during the projection. In VISAT, an object is usually visible 
in 3-4 image stereo pairs. Consequently, the focusing area 
(processing window) of object detection and verification on the 
images is located (shown in Figure 4). 
The module of verification of the existence of objects is to identify 
whether or not the object exists (or misses). It involves a number of 
processing steps: detection of vertical edges, formation of line 
segments and feature correspondence of line segments. Once the 
existence of the object is verified, the module of multinocular line 
reconstruction is executed to determine the accurate position of the 
object, and then update the position information of the object in its 
database. 
It has been shown that this approach is able to detect the desired 
features at a probability of about 85% and the efficiency is 
about as twice as that of manual processing. The advantage of 
this approach is that the use of prior information of the object 
positions greatly simplifies the procedure of object detection 
and makes the processing results much reliable. This approach 
has gained a great attention of the user community. Requests for 
more tests and licensing of the software are being processed. 
4.2 Image Bridging 
GPS positioning techniques are particularly susceptible to cycle 
slips, multi-path, and other degrading effects during data 
collection. To overcome these outages, the high short-term 
accuracy of the INS may be exploited to bridge the GPS-derived 
vehicle trajectory. However, the length of the bridging interval 
is limited by the quality of the INS, as errors in the INS tend to 
accumulate rapidly when operating unaided. 
Practical experience has shown that, in certain cases, the 
integrated GPS/INS component may not be able to effectively 
bridge the GPS outages in the post-processing stage. It is also 
possible that the GPS or INS component may fail completely, 
due to power failure or other logistical reasons. In the former 
case, the system accuracy may degrade beyond some threshold 
value, and may no longer meet accuracy requirements. In the 
latter (and worst) case, the stereo imagery potentially cannot be 
georeferenced, and system functionality is lost. Current research 
work aims at the development of an alternative image-based 
bridging strategy, thereby enabling the estimation of the three- 
dimensional motion parameters of the system from the sequence 
of time-varying stereo imagery. Since the strategy is solely 
image-based, it operates in the absence of both GPS and INS 
data, if required. It is, therefore, applicable when the vehicle 
trajectory cannot be bridged using INS data. 
A feature-based approach to the motion estimation problem is 
adopted, based on four primary steps: 
• extraction of candidate features from each image of two 
adjacent stereo pairs independently; 
• establishment of stereo (spatial) correspondence of 
candidate features within each stereo pair; 
• establishment of temporal (motion) correspondence of 
candidate features between adjacent stereo pairs; and 
• determination of the orientation parameters of the 
subsequent stereopairs relative to the first from the 
corresponding (conjugate) features. 
The so-called correspondence problem has long been 
recognized as a fundamental problem in multi-view image 
analysis. In stereo image sequence analysis, both stereo (spatial) 
and temporal (motion) correspondence problems must be 
overcome in order to estimate motion parameters. Therefore, 
steps 2 and 3 are the critical steps in motion estimation 
algorithms. In this work, the stereo correspondence analysis is 
area-based, driven by the extracted interest points. The strategy 
proceeds in two stages: initial correspondence and precise 
registration. 
During the initial correspondence, feature correspondences are 
hypothesized using a weighted correlation-based similarity 
measure, with the search constrained by the epipolar constraint. 
The weighted correlation measure reduces the influence of 
perspective distortion prevalent in terrestrial imagery. The list 
of candidate features is pruned using a disparity range 
constraint; the role of the images is then reversed, and candidate 
matches that satisfy the left-to-right consistency constraint are 
retained as correct. The images, therefore, play a symmetric role 
in the matching process. Final precise registration of conjugate 
points is achieved by least squares matching (Gruen, 1985). 
Accurate sensor calibration allows the computation of 3D 
coordinates for all (hypothesized) conjugate features, based 
upon initial estimates of the six parameters of exterior 
orientation derived from a model of the vehicle motion. The 
static stereopsis is, therefore, used to aid the recovery of 
temporal correspondence. Constraints are incorporated to 
constrain the relative imaging geometry, and control the 
propagation of errors. 
2-5-7