Improved Image Analysis Technique for Embedded System
Abstract:
Embedded system design is a challenge especially when devices have limited memory.
For these systems, comparing two images is really computationally expensive.
Storing the whole image’s information and then compare these two images is a
big deal. Although logarithmic algorithm gives us a better and computationally
inexpensive technique for matching a small picture in a large scene, we need to
match the whole image and for embedded system which uses microcontroller for
all computational works it is almost impossible. Because microcontroller has
limited number of registers and its size of EEPROM and programmable memory is
very small. Our proposal is to match and differentiate two images based on some
specific features in a specific position of the object. Clustering algorithm
clusters particles, objects in a specific group. To differentiate two objects
of a specific group we again need to match two objects wholly. Without matching
two objects wholly, we can find out some unique feature in the objects and by
comparing their positional existence we can distinguish two objects and
simultaneously reduce the computational complexity.
Keywords:
Embedded System, Cluster, Feature Set.
Introduction:
According to [Huba 82], presence or absence of any feature can be used to
identify an animal’s genus group. For example, presence or absence of lungs can
be used to identify the category of the mammal, whether the mammal lives in sea
or in land. Clustering algorithms can also be used to cluster the mammals [Jain
88]. Now, if we can find out a specific feature for all mammal lives in the sea
we can use that feature’s positional existence or size to distinguish two
mammals. For our test we use two insects from arthropod, one is cockroach and
another is spider. Both the cockroach and spider contain a sensory element in
front of their head. For cockroach its length is almost 2 inches, but for spider
it is almost half inches. It is a unique feature to distinguish these two
objects for matching. We did not do any edit distance [Theo 03] or difference
calculation for matching. Matching is simply based on the color components of
the pixel in an image.
Contribution:
Matching algorithm is tested in MATLAB for correct matching. It is also tested
in the MPLAB’s simulator for testing using Programmable Interface Controller
(PIC) microcontroller. This mechanism is adopted for designing a protection
device. The objective behind the device is the security of human being. Most of
us hate insects like cockroach, spider. The protection device, namely,
“Environment Aware protection Device” is targeted to sense the environment by
capturing the image of the insects, detecting the hazardous insects using this
matching algorithm for awarding about the presence of an insect.
References:
[Huba 82] Hubalek Z.
“Coefficients of association and similarity based on binary (presence-absence)
data-an evaluation,” Biological Review, Vol. 57, pp.669-689, 1982
[Jain 88] Jain A.K.,
Dubes R.C. Algorithms for Clustering Data, Prentice Hall, 1988.
[Theo 03] Theodoridis
S., Koutroumbas K. Pattern Recognition, Academic Press, 2003.
Comments
Post a Comment