Evaluation of feature extraction techniques on aquaculture place recognition problem

Aquaculture industry is one of the most sustainable and environmental friendly way of satisfying the world’s increasing food demand. However, fish escape still remains a critical challenge that is not only a financial loss but badly effects the biodiversity as well. These escapes mostly happens through a hole caused by a natural event or an accident hence it is essential to have a mechanism to detect it. SLAM(simultaneous localization and mapping) is the current state of the art method to resolve issues like this. The aim of this thesis is to focus on the Place re- cognition and loop closure part of the SLAM. BOW technique along with different feature extraction methods will be evaluated and compared in different realistic scenarios. Moreover, the performance of these techniques on non net cage marine data will be also discussed to establish a strong argument.