A
Combined Approach for Stereoscopic 3D Reconstruction Model based on Improved
Semi Global Matching
Rajeshkannan Sundararajan and Reeba Korah
Department of Electronics Communication and Engineering, St. Joseph’s College of Engineering, India
Abstract: The effective recovery of the 3D structure of a scene using two
or more 2D images of the scene, each acquired from a different viewpoint is a
challenging task of stereovision. Defining pixel correspondence in stereo pairs
is a fundamental process for automated image based effective 3D reconstruction.
This paper presents modified Census based approach for local cost optimization
where local matching cost is combined with Sum of Squared Absolute Differences
(SSAD) of the image color values and then aggregated. From the aggregated cost,
effective disparity map is obtained using Semi Global Matching (SGM) which
improves the quality of the matches. This proposed approach represents a fusion
of state of the art algorithms to improve the matching quality with reduced
number of bad pixels. Finally, a stereoscopic 3D view will be obtained by
merging triangulation algorithm in the realistic manner. Because of more
realistic depth perception, our proposed three dimensional stereo model finds
application in medical field where it improves surgical success with shorter
operation time and research in space where effective analysis can be made by
using calibrated photo realistic 3D model of the space structure.
Keywords: Cost function, disparity map, modified census transform, SAD, SGM, SSAD, triangulation.
Received August 25, 2013; accepted March 14, 2014