Page 1 of 7
Journal for Studies in Management and Planning
Available at
http://edupediapublications.org/journals/index.php/JSMaP/
e-ISSN: 2395-0463
Volume 02 Issue 12
December 2016
Available online: http://edupediapublications.org/journals/index.php/JSMaP/ P a g e | 31
Reconstruction Depth of scattered samples
Representation, algorithm, sampling
1 Sundaragiri Raju,2 Mrs.G.Satya Prabha
1PG Scholar, Department of ECE, SLC's Institute of Engineering and Technology, Piglipur Village, Hayathnagar
Mandal, Near Ramoji Film City, Ranga Reddy District, Hyderabad, Telangana
2Assistant Professor, Department of ECE, SLC's Institute of Engineering and Technology, Piglipur Village, Hayathnagar Mandal,
Near Ramoji Film City, Ranga Reddy District, Hyderabad, Telangana
Abstract The rapid development of 3D
technology and computer vision applications
has motivated a thrust of methodologies for
depth acquisition and estimation. However,
existing hardware and software acquisition
methods have limited performance due to
poor depth precision, low resolution, and
high computational cost. In this paper, we
present a computationally efficient method
to estimate dense depth maps from sparse
measurements. There are three main
contributions. First, we provide empirical
evidence that depth maps can be encoded
much more sparsely than natural images
using common dictionaries, such as wavelets
and contourlets. We also show that a
combined wavelet–contourlet dictionary
achieves better performance than using
either dictionary alone. Second, we propose
an alternating direction method of
multipliers (ADMM) for depth map
reconstruction. A multiscale warm start
procedure is proposed to speed up the
convergence. Third, we propose a two-stage
randomized sampling scheme to optimally
choose the sampling locations, thus
maximizing the reconstruction performance
for a given sampling budget. Experimental
results show that the proposed method
produces high-quality dense depth
estimates, and is robust to noisy
measurements. Applications to real data in
stereo matching are demonstrated.
I. INTRODUCTION
THE rapid development of 3D technology
has created a new wave of visualization and
sensing impacts to the digital signal
processing community. From remote
sensing to preserving historical and from
rescue to 3D laparoscopic surgery the
footprints of 3D have been influencing a
broad spectrum of the technological
frontiers. This paper follows the concept of
reproducible research. All the results and
Page 2 of 7
Available online: http://edupediapublications.org/journals/index.php/JSMaP/ P a g e | 32
examples presented in the paper are
reproducible using the code and images
The purpose of this supplementary report is
to present experimental results on the
selecting parameters for the proposed depth
reconstruction algorithm. The total size of
the file is 769 kB. Contact l7liu@ucsd.edu
for further questions about this work. Color
versions of one or more of the figures in this
paper are The successful development of
3D signal processing isfundamentally linked
to a system’s ability to acquire depth. To
date, there are two major classes of depth
acquisition techniques: hardware solutions
and computational procedures. Hardware
devices are usually equipped with active
sensors such as time-of-flight (ToF) camera]
and LiDAR While being able to produce
high quality depth maps, these hardware
systems have high instrumentation cost.
Moreover, the data acquisition time of the
devices is long (e.g., a recently proposed
ToF cameras can only achieve 10fps [8],
whereas standard cameras nowadays can
easily achieve 60fps.) Although speeding up
is possible, spatial resolution has to be
traded off in return. An alternative solution
to acquiring depth is to estimate depth using
a set of computational procedures. This class
of computational methods, broadly referred
to as disparity estimation algorithms
estimates the depth by computing the
disparities between a pair of stereo images
through their corresponding matching points
Disparity estimation algorithms usually
work well under well conditioned
environments, but they could be sensitive to
illumination, noise, stereo alignments, and
other camera factors. Thus, the effective
number of matching points that one can use
for disparity estimation is actually much
fewer than the number of pixels of the depth
map
A. Scope and Contributions
The objective of this paper is to present a
sampling and reconstruction framework to
improve and speed up the depth acquisition
process. The key idea is to carefully select a
sparse subset of spatial samples and use an
optimization algorithm to reconstruct the
final dense depth map.
II. NOTATIONS AND PROBLEM
FORMULATION
In this section we introduce notations and
elaborate on the problem formulation.
A. Depth and Disparity
The type of data that we are interested in
studying is the depth map. Depth can be
directly measured using active sensors, or
inferred from the disparity of a pair of stereo
images. Since the correspondence between
depth and disparity is unique by simple
Page 3 of 7
Available online: http://edupediapublications.org/journals/index.php/JSMaP/ P a g e | 33
geometry [29], in the rest of the paper we
shall use depth and disparity
interchangeably.
B. Sampling Model
From remote sensing to preserving historical
and from rescue to 3D laparoscopic surgery
the footprints of 3D have been influencing a
broad spectrum of the technological
frontiers. This paper follows the concept of
reproducible research.
C. Penalty Functions
As discussed is not an effective formulation
because the _1 norm penalizes both the
approximation (lowpass) and the detailed
(highpass) coefficients. In reality, since
disparity maps are mostly piecewise linear
functions, the lowpass coefficients should be
maintained whereas the highpass
coefficients are desirable to be sparse. To
this end, we introduce a binary diagonal
matrix W ∈ RM×M where the ( j, j )th entry
is 0 if j is an index in the lowest passband,
and is 1 otherwise. Consequently, we
modify the optimization problem as Finally,
it is desirable to further enforce smoothness
of the reconstructed disparity map.
Therefore, we introduce a total variation
penalty so that the problem becomes
minimize
III. SPARSE REPRESENTATION OF
DISPARITY MAP
The choice of the dictionary _ in (11) is an
important factor for the reconstruction
performance. In this section we discuss the
general representation problem of disparity
maps. We show that disparity maps can be
represented more sparsely than natural
images. We also show that a combined
wavelet contourlet dictionary is more
effective in representing disparity maps than
using the wavelet dictionary alone.
A. Natural Images vs Depth Data
Seeking effective representations for natural
images is a well-studied subject in image
processing However, representations of
disparity maps seem to be less studied. Fig.
1. PSNR values of approximating a disparity
patch and a image patch using the leading
