 This paper proposes a novel approach to classifying infant body movements based on the general movement assessment, GMA. It introduces a set of new features and a feature fusion pipeline to improve the accuracy of the classification task. Additionally, it evaluates the performance of the proposed method against existing approaches using two different datasets. The experimental results demonstrate that the proposed method outperforms existing approaches in terms of accuracy and generalization ability. Furthermore, the proposed features provide more detailed information about the infant's body movements, allowing for better interpretation of the results. This article was authored by Kevin D. McKay, Peng Peng Hu, Hubert P. H. Shum and others.