J.L. Baker, D. Campbell, A. Bodnarova, and V. Chandran (Australia)
Robotic Sensing, Data Fusion, Robotic Vision, Colour Seg
This paper presents a novel approach to segment based
stereo correspondence. The method presented utilises
colour information to determine segments of an image
that represent physical objects. The stereo correspondence
framework proposed is based upon using the colour seg
ments for feature matching across stereo images. The
colour segments are determined using a unique hue
intensity segmentation fusion algorithm. The segmentation
algorithm can be expanded to include motion information.
Results illustrate the success in combining hue and inten
sity information to locate segments representative of phys
ical objects, useful for performing stereo correspondence.
The approach proposed in this paper is a hybrid area
feature based approach. Initially the extracted object seg
ment boundaries are mapped between the left and right
stereo images. The result of the simple and easier segment
based stereo correspondence is utilised as a constraint in
the search space of the area based approach. The novel as
pect of this algorithm is the use of segments as the basic
feature for feature matching.