4
complexity and variety of microstructure, most traditional
methods could not balance speed and accuracy of stitching
strategy. To overcome this problem, we develop a method
named very fast sequential micrograph stitching (VFSMS.
Thus, the offset along the vertical and horizontal directions
as shown in Figure 5 can be calculated from the matched
key points to represent the transformation between two
adjacent borehole images. Starting from the first frame,
these offset values can be calculated one by one for the fol-
lowing sequential frames until the last frame and are stored
as an offset list.
Image alignment involve aligning one image to another
and then warping the images onto the final canvas to cre-
ate a panoramic borehole image. After obtaining the list
of video frame and offset, starting from the first one, the
frames in the video frame list can be aligned one by one
based on the calculated offset values in the offset list to
generate the panoramic borehole image. If necessary, image
blending can be conducted to create a seamless transition
between the overlapped regions to ensure the visual consis-
tency of the stitched images without noticeable seams.
However, the sequential image stitching process may
fail at some point. When the algorithms cannot detect
enough key points or cannot find good matches, the offset
between two adjacent frames cannot be estimated, lead-
ing to the failure of the image stitching process. In order
to continue the stitching process, the new frame is moved
to the bottom of the previous frame and is further moved
down with a gap to show the location of stitching failure.
As a result, a gap in the stitched borehole image indicates
the disconnection of the following image segment from the
previously stitched segment.
BOREHOLE VIDEOS WITH DIFFERENT
BOREHOLE CONDITIONS
It can be found from the image stitching process in Section
2 that the detection and matching of key points between
overlapped video frames are crucial for the sequential image
stitching. The stitching process would fail if the algorithm
cannot find enough key points in the borehole images or
cannot find good matches for the key points in two adja-
cent video frames. The appearance of a borehole may vary
with the wall condition. With a freshly drilled borehole,
the wall is normally covered with rock dust. If a borehole is
freshly washed, water remains on the wall and reflects light.
These wall conditions can potentially affect the process of
key point detection and matching and further affect the
generation of panoramic borehole images. One objective
of this study is to investigate the influence of borehole wall
conditions through the performance of the image stitching
methodology.
The same borehole was scoped three times under dif-
ferent conditions and three borehole videos were recorded
for this study. As shown in Figure 6, the IPLEX GX video
borescope from Olympus America was used (Olympus).
The borehole was in the roof of a room-and-pillar mine
in West Virginia, USA. The diameter of the borehole was
25.4 mm (1 inch). The depth of the borehole was measured
to be 2.56 m with very limited roof deformation during
the study period. The videos were recorded when the side-
view borescope tip moved downward towards the roofline.
The moving speed and orientation were maintained at a
constant rate as much as possible during scoping. The first
video was recorded on 12/6/2023, a few weeks after the
development of the entry, and there were rock dust cover-
ing the borehole wall. The second video was recorded on
1/5/2024 after the borehole was freshly washed with water
to remove the rock dust from the borehole wall. The third
video was recorded on 1/17/2024, about 12 days after the
washing when the borehole wall was assumed to be dry. The
Figure 5. Offset calculation between two adjacent images
Figure 6. The side-view video scope used in this study
complexity and variety of microstructure, most traditional
methods could not balance speed and accuracy of stitching
strategy. To overcome this problem, we develop a method
named very fast sequential micrograph stitching (VFSMS.
Thus, the offset along the vertical and horizontal directions
as shown in Figure 5 can be calculated from the matched
key points to represent the transformation between two
adjacent borehole images. Starting from the first frame,
these offset values can be calculated one by one for the fol-
lowing sequential frames until the last frame and are stored
as an offset list.
Image alignment involve aligning one image to another
and then warping the images onto the final canvas to cre-
ate a panoramic borehole image. After obtaining the list
of video frame and offset, starting from the first one, the
frames in the video frame list can be aligned one by one
based on the calculated offset values in the offset list to
generate the panoramic borehole image. If necessary, image
blending can be conducted to create a seamless transition
between the overlapped regions to ensure the visual consis-
tency of the stitched images without noticeable seams.
However, the sequential image stitching process may
fail at some point. When the algorithms cannot detect
enough key points or cannot find good matches, the offset
between two adjacent frames cannot be estimated, lead-
ing to the failure of the image stitching process. In order
to continue the stitching process, the new frame is moved
to the bottom of the previous frame and is further moved
down with a gap to show the location of stitching failure.
As a result, a gap in the stitched borehole image indicates
the disconnection of the following image segment from the
previously stitched segment.
BOREHOLE VIDEOS WITH DIFFERENT
BOREHOLE CONDITIONS
It can be found from the image stitching process in Section
2 that the detection and matching of key points between
overlapped video frames are crucial for the sequential image
stitching. The stitching process would fail if the algorithm
cannot find enough key points in the borehole images or
cannot find good matches for the key points in two adja-
cent video frames. The appearance of a borehole may vary
with the wall condition. With a freshly drilled borehole,
the wall is normally covered with rock dust. If a borehole is
freshly washed, water remains on the wall and reflects light.
These wall conditions can potentially affect the process of
key point detection and matching and further affect the
generation of panoramic borehole images. One objective
of this study is to investigate the influence of borehole wall
conditions through the performance of the image stitching
methodology.
The same borehole was scoped three times under dif-
ferent conditions and three borehole videos were recorded
for this study. As shown in Figure 6, the IPLEX GX video
borescope from Olympus America was used (Olympus).
The borehole was in the roof of a room-and-pillar mine
in West Virginia, USA. The diameter of the borehole was
25.4 mm (1 inch). The depth of the borehole was measured
to be 2.56 m with very limited roof deformation during
the study period. The videos were recorded when the side-
view borescope tip moved downward towards the roofline.
The moving speed and orientation were maintained at a
constant rate as much as possible during scoping. The first
video was recorded on 12/6/2023, a few weeks after the
development of the entry, and there were rock dust cover-
ing the borehole wall. The second video was recorded on
1/5/2024 after the borehole was freshly washed with water
to remove the rock dust from the borehole wall. The third
video was recorded on 1/17/2024, about 12 days after the
washing when the borehole wall was assumed to be dry. The
Figure 5. Offset calculation between two adjacent images
Figure 6. The side-view video scope used in this study