You are on page 1of 3

1 Topic

The goal of this paper is to create an overview of digital image processing in the
field of auditory acoustics. The subject is a device (Perhaps a moving robot)
that is capable of finding its way around close quarter mazes or labyrinths using
ultrasonic scanning of the surroundings at evenly spaced intervals of θ.

2 Recording Information
The use of devices designed for high precision such as stepper motors are ideal
for the recording of nformation reliably, however for the purpose of this as-
sessment normal DC brushed DC motors were used with constant voltage and
careful timing. The content in this document assumes that the recording device
moves in full 360 degree motion. In order to prepare memory for recording, the
resolution of the desired recorded information must be known. The high the
resolution of recorded information, the more memory will be required. Fortu-
nately, the resulting information only needs an single dimensional array of cells
and therefore grows linearly. For most micro controller applications, available
memory should never be a problem. This document will use a resolution of 16
samples per scan or 16 cells that cover 360 degrees. every 8 cells is 180 degrees
and every four is 90.
Once a destination cellular matrix is ready to buffer scan information, the
scan should commence and rotate in accordance to the timing that allows both
for accurate ultrasonic response, but also precise timing. A low resolution scan
of 16 samples may be recorded like the following:

12 43 49 30 5 4 5 200 255 255 130 10 3 8 14 24

3 Convolution Masks
Because the recorded scan of an ultrasonic sweep can be recorded as a 1 pixel
high image matrix, the neighbors of any ”pixel” can be used in conjunction with
a convolution mask in order to perform certain types of filtering or enhancement
to the recorded information. Because the resulting set of data is only 1 pixel
tall, the result of convolution masks may perhaps be limited when for example
searching for vertical lines or anything that might require information from
vertically-offset neighbors.
A centered, single dimension convolution mask applied to a two dimensional
scan of ultrasonic values, the first of which the coordinate x represented as angle
θ, and the second a recorded value of the scan at θ, can be performed using the
following equation:

1
 
θ+β
X
ω= f (x) · g(θ + β + α − x) {dm <= ω <= dM }
x=θ−β

Where ω is the new resulting value at angle θ. θ − β and θ + β represent the


beginning end of the image kernel. g(x) represents the convolution mask itself,
with θ + β + α − x passed in order to align the beginning of the convolution
mask access to 0. The output domain is represented from domain minimum dm
to domain maximum dM where their values can be from zero to the maximum
possible value output of the recorded information. In the event that the values
fall below or above the output domain of ω, they should be truncated either at
dm or dM respectively. β represents 1 less than half of the size of the convolution
mask in order to find the pixel coordnate in the resulting ultrasonic scan, and
can be defined as:
α−1
β=
2
Where α = the size of the convolution mask. Therefore, if the size of the
convolution mask is 5, then there will be two values to the left of θ, one centered
at θ, and two to the right of θ

4 Enhancement of ultrasonic contrast through


histogram equalization
Histogram equalization can be used to enhance the overall distribution of con-
trast in recorded ultrasonic information with the intention of improving the
reliability of search algorithms such as A*. Creating a histogram and perform-
ing equalization is the same procedure as used in digital image processing. The
histogram can be formed by creating an array of the occurrence of equivalent
values, for each value of θ. Then, perform equalization by integration:
Z
h(x)dx

Where h(x) is the histogram matrix of the ultrasonic scan. Then, normalize
the output of the histogram within its appropriate maximum domain in order
to finalize the histogram equalization for the ultrasonic scan.

5 Path finding / searching by use of the A* al-


gorithm
After histogram equalization and any other preparation to the information
recorded from the ultrasonic scan is completed, the information obtained can be
used in conjunction with the A* algorithm in order to solve simple tasks such

2
as working through a maze using the A* algorithm. This is easiest given that
the machine a) has a relative idea of where it is, and b) has a relative idea of
where it wants to go.
The information processed from the result of the ultrasonic module (such as
distance in centimeters) will grow higher the farther away its destination object
is, until it completely falls off and the distance becomes undefined or infinite,
depending on semantics used. There are a few problems with using ultrasonic
information to process A* searching. Namely, coming up with a solution for
cost. It may be easier to interpet the values at f (θ) as inverted, since a wall
should be interpreted as a move of extremely high cost, or not an option of
movement, depending on implementation and scenario. This becomes more
complicated, since the procedure mat or may not want to consider variable
lengths of movement. In grid movement, it becomes simple, but in variable
movement, a forward movement threshold should be calculated a head of time
per move to ensure that the device is always moving the optimal range for best
results. This could be achieved by calculating g(x) in the a* algorithm set of
variables at multiple points, then finding the point in which is works best, and
considering that location as a candidate successor.
Inconsistent results read from the ultrasonic device can throw off the relia-
bility of he searching procedure. The likeliness of this is reduced by applying
filters to reduce noise and perhaps smoothing the resulting image of the ultra-
sonic scan. Once error levels are below a particular threshold, the device can
also use the image to determine the appropriate angle of movement, unless the
device moves orthogonally. By using basic trigonometry, the device can keep
track of its current position relative to orthogonal coordinates to the finish for
calculation of h(n) ( The remaining estimate of distance to the goal).

6 Conclusions
Digital Image processing can apply to more than just images. It should work
on any model of information that can represent any form of matrix. In the
case of ultrasonic scans in one angle of direction, a single pixel tall, multiple
pixel wide image is created that could even be converted to a grayscale photo
strip for visual observation. Image processing techniques work because they are
designed to perform operations on data. The photograph used is nothing more
than photographic input that is easily related to a human being who looks at
this form of information on a daily basis. If photographs ca be converted to
information usable in mathematics, then the opposite is also true.

You might also like