MOVING OBJECT DETECTION WITH MEMRISTIVE CROSSBAR ARRAYS
Loading...
Date
2016
Authors
Duisenbay, Sultan
Journal Title
Journal ISSN
Volume Title
Publisher
Nazarbayev University School of Engineering and Digital Sciences
Abstract
This thesis is dedicated to the hardware implementation of a novel moving
object detection algorithm. Proposed circuit includes several stages, each of
which implements a particular step of the algorithm. Four higher bit planes are
extracted from a grayscale image and stored in memristive crossbar arrays, and
the respective bit planes are compared via memristive threshold logic gates in
XOR configuration. In the next stage, compared bit planes are combined by
weighted summation, with a highest weight assigned to MSB plane and smaller
weights for less significant bit planes. After summation stage, obtained grayscale
image is thresholded to obtain binary image. The last stage is implemented via
memristive content-addressable memory array, which serves two purposes. It
is used as a long-term memory in comparison to crossbar arrays, which serve
as a short-term memory of proposed circuit. Content-addressable memory
is updated based on the row-by-row difference between first and second pair
of frames processed by previous stages. It also allows for analysis of object
movement direction and velocity by observing the row capacitors’ discharge.
Simulations show that accuracy of proposed circuit operation is increased with
the larger array size. Delay analysis of the circuit is carried out, power and area
calculations show that proposed circuit is a viable candidate as a co-processing
operator for existing image sensors.
Description
Keywords
object detection algorithm
Citation
Sultan Duisenbay; 2016; MOVING OBJECT DETECTION WITH MEMRISTIVE CROSSBAR ARRAYS; School of Engineering. Department of Electrical and Electronic Engineering. Nazarbayev University; http://nur.nu.edu.kz/handle/123456789/2316