A Knowledge-Based Planner for Processing Unconstrained Underwater Videos

Video data collected continuously are pervasive today but analyzing them in an efficient manner has proven to be a challenge. This is because raw data is unlabelled and prone to noise, causing difficulty in extracting knowledge. With the aid of userprovided domain knowledge and heuristics used by image processing experts, an automated solution is implemented. It makes use of formalisms for goaldirected behavior in the form of hierarchical task networks (HTNs). These are incorporated within a novel workflow composition framework that aims to assist naive users conduct complex video processing tasks automatically. An example is illustrated for video classification, fish detection and fish counting in unconstrained underwater videos.