Human Action Recognition Based on Body Shape and Orbicular Grid

Proliferation of online images, the recognition of human actions within still images has emerged as a prevalent research area within the realm of computer vision. The subject pertains to the identification and categorization of human actions captured within static images. A novel approach is introduced in the study which involves the utilization of structural features in conjunction with an orbicular grid encompassing the human–object interaction. By modifying grid parameters and employing various classifiers, experiments were conducted and assessed using benchmark datasets. The proposed method showcased impressive accuracy exceeding 90% across numerous actions. Furthermore, a comparison with a recent semantic-based approach highlighted the heightened significance of spatial relationships in the context of human action recognition.