Performing visual feature extraction in a network of processing
nodes is challenging and requires the development of algorithms
that perform the allocation and delegation of computational tasks,
of protocols responsible for the correct transmission of data, etc.
This report starts working from the implementation of a testbed
for the evaluation of distributed processing of visual features and
extends its functionality. The testbed is implemented in C++ us-
ing credit-card sized computers and ZigBee USB units. Communi-
cation between nodes is carried out through ASN.1 defined types.
The first part of the thesis work consists of enabling the system
to work with two different feature extraction schemes. The first
one is the already implemented SURF algorithm from OpenCV
and the second one is the original implementation of BRISK. The
user must be capable of choosing indifferently between them. The
second part of the work consists of modifying the transmission
aspects of the system including the necessary classes to provide
reliability via the design of a retransmission protocol. The reliable
transmission protocol used is a version of a Stop-and-Wait scheme.
The system’s performance is evaluated detailing the time needed
to complete each step of the feature extraction process, presenting
a comparison between the SURF and BRISK detection and ex-
traction times, and computing the frame loss rate and achievable
throughput once the retransmission protocol is implemented.