Nowadays, the volume of video content is on a rapid incline while a fundamental need appears to extract relevant information without human intervention.
In this paper I deal with a system being fit for object tracking in videos and capable of classification of the tracked objects. Both tracking and classification are complex and difficult tasks and up to date lots of research projects are run linked to them.
My thesis will present the system being able to track underwater creatures while identifying its species. Based on the tracking and the classification results my system creates statistics on the actual fish species.
In this thesis I review the state-of-the-art methods in object tracking such as Kalman-filter and the background subtraction as well as the ones being the base of my classification subsystem like the SURF descriptor, the C-SVM classifier and the Bag-of-Words model.
The system presented by this paper is built on the dataset published during a competition organized by ImageCLEF organization. As a consequence the system was trained to recognize 15 different fish species, adding to that also the system test itself was made on the same dataset.