Method of Focus of Expansion (FOE) is proposed by 29. In matching
between images, FOE is estimated by hand and Harris operator is employed to
detect the feature points. Next, the images are projected in the direction to
side-view applying the projective transformation. This formed the panoramic
image on the basis of the image correspondences. After all, the difficulties in
the transformation of front-view image to side-view image, causing the image is
not entirely anterior view to the buildings.
A paradigm, Random Sample Consensus (RANSAC) is
introduced by 30 for fitting a
model to experimental data. RANSAC has the capability to interpret and smoothen
the data which contains a significant percentage of gross error. RANSAC is
suitable for the applications of automated image analysis which target to
interpret on the data provided by error-plane feature detectors.
to RANSAC requirement, derivation of the results is based on the minimum number
of landmarks needed in getting a solution. Then, an algorithm is proposed for
computing these minimum landmark solutions in closed form. Even if the
landmarks provided are incorrect located due to terrain changes or low
visibility, the system still able to perform. In order to obtain an initial
solution and eliminate the invalid data points, RANSAC utilize as small an
initial data set as feasible. It also enlarges the set with consistent data
31 performed multiple-step procedure which utilizes
geometric constraints which aims to reduce the probability of false
correspondences in RANSAC approach. With that, the initial translation and
rotation of the robot is set to zero. From the known maximum translational and
rotational velocity of the robot, the uncertainty of the change in robot’s pose
can be estimated.
process of image alignment where image(s) may have incorrect is called the
image reprojection. This process is performed in order to correct the geometric
distortions 23. The distortions is often caused by the lenses of
camera which used to capture the images. Image reprojection process is also
referred as image undistortion. Image warping is performed as well to transform
the image from planar projection surface to spherical projection surface.
Figure 2.12 shows the geometric relationship between two different image representations.
The 2D Cartesian coordinates is denoted by (x, y) while spherical coordinates
denoted as (?, ?), in which, 0 ? ?