SURF-FAST BASED PARTICLE FILTER TRACKING OF ARAPAIMA GIGAS

Abstract

Invasive Alien Species (IAS) of fish has recently become issue of concern, due to their
adverse ecological effect, as well as potential risk and danger to humans. Method of
containment mostly employed to invasive fish without harming indigenous fish species
would involve direct human effort. The involvement of humans in physical and direct
containment of invasive fish species can be very tedious, as it involves diving and
hunting of alien fish species. However, the use of vision based underwater robots can
greatly reduce the cost, effort and risk involved, as well as yield more result in shorter
time. Underwater robot vision system is primarily built upon visual recognition and
tracking. Due to the nature of underwater environment, as well as tracking target, it
becomes necessary that the underwater tracker should have good performance. In this
study, the particle filter tracking algorithm is employed for underwater tracking of
Arapaima Gigas, where modifications for its improvement were proposed. The
improvement is towards enhancing the tracker performance in terms of accuracy and
tracking error using multi-likelihood of different tracking features. The features used
for tracking are Speeded Up Robust Feature (SURF) and Fast Accelerated Segment
Test (FAST). The result from multi-likelihood SURF-FAST tracker was the better than
single feature FAST or SURF trackers in terms of performance indices, namely
accuracy and tracking error. However, better performance can be achieved when
implemented on a graphics processor, also the tracker needs to be validated inside a
real underwater environment.