|
Original picture image
Extraction of edges
Detection of parallel lines
Making of an environmental map
|
|
Analysis of space structure from a scene image
We are researching the analysis of a space structure from a scene
image from an ocellus aspect to achieve creation of an autonomous
mobile robot. In cases of space recognition by stereo vision, we
detected the correspondence point from two images with two or more
cameras, and the parallax was analyzed and the distance was measured.
However, in cases of the ocellus aspect, since neither the correspondence
point nor the parallax is utilized, we use a line information on
the edge (outline) of the object within the plane image. Then, we
analyzed the space by using the characteristics of a "disappearance
point" which is where the extensions of two parallel lines in a
three-dimensional space always intersect by one point when projected
onto a plane.
First, the Sobel filter is used on the original picture image
so that the edges can be extracted. Next, the segments are continuously
emphasized, supplemented and integrated. In addition, an extra segment
is removed. Moreover, the segments of which the extension line intersects
by one point (disappearance point) are detected as parallel lines.
Finally, space compositions such as the floors, ceiling, walls,
and corners are specified from the intersection of the parallel
lines and the point of contact, etc., and an environmental map is
created.
The aforementioned processing was programmed utilizing C language,
and we experimented on the computer with 40 images of a room which
were taken beforehand. As a result, the absolute coordinates were
not necessary for this technique, like strict space analysis by
stereo viewing. However, when we presume the range of paths needed
to pass, we confirmed that it was effective in rooms where a lot
of parallel lines exist.
|