Footballer Detection on Position Based Classification Recognition using Deep Learning Approach

In the current state of the development of healthy project of mass entertainment, sports events are more often presented in the forms of videos, which are concerned and discussed by public. However, the accuracy of the information is inadequate as it is beyond the human brain. Therefore, this is where computer technology will be getting involved as it has been increasingly applied to each sector of human society, extending the human brain and perception abilities including detection and tracking a right position of a target in sport videos. Hence, automated analysis technology of sports videos is such a demand especially for football game videos with extremely wide audience. This paper proposed an automatic approach to identify players on position based for post-match evaluation for both coaches and players. We adopt a publicly available dataset, IAUFD to train and evaluate the proposed framework. First, a video or image will be input to a split feature extraction and matching by implementing ResNet50. This purposely for extracting a player from occlusion scenarios. Next, it will be further input into a deep learning algorithm, VGG-M CNN to accurately recognise a player. This approach perhaps will improve a football team’s tactical and increase individual performance. As a results, we achieve validation accuracy of 97.14%. In future, a details evaluation of players performance can be achieved with an automatic visualisation which is convenient to the users.