Sports action recognition and event Captioning system for videos using Heuristics

Abstract

Computer vision is a multidisciplinary scientific field that allows computers to gain a high level understanding of digital images and videos. It is done in the perspective of automating the task of human visual system. Major tasks in computer vision include acquisition, cleaning, analyzing and extraction of high-dimensional data in the form of decisions from the real world. Various essential fields such as agriculture, healthcare, sports, transportation, manufacturing, retail and so on make use of computer vision-based approaches to provide the sense of sight for machines. Among these domains, sports is one of the fields that characterises fast and swift motion which cannot be followed and understood through naked eyes within a short period of time. Hence, computer vision-based approaches are adopted in the field of sports in various aspects such as training players, fitness maintenance, dynamic decision making during live events and so on. To make decisions on such critical situations, computer vision-based approaches use the knowledge acquired through continuous learning. The knowledge acquired through context of the content at every processing stage with respect to the key element is termed as concept. Therefore, the focus of the proposed research work is to provide conceptual captions automatically, by linking all the key contexts that satisfy the criteria of high prominent thereby handling challenges in manual captioning that may generate non-conceptual captions. As sports videos consist of various sequences of motion (jumping, throwing, ball rolls on, audience cheering and so on), the key element is considered to be motion. newline