2025 Volume 6 Issue 1 Pages 384-394
In recent years, an increasing number of companies have adopted digital technology with the aim of reducing work losses. However, the implementation of such technology is frequently associated with considerable expenses and labour requirements. This study proposes a methodology for mitigating the challenges associated with data collection by employing action classification through the analysis of skeletal information extracted from first-person and third-person view videos.Conventional methods encounter two primary issues in action classification using still images: the absence of information on actions such as walking, and the paucity of information on work objects and the environment provided by skeletal information. The proposed methodology is a two-stage model that aims to address these shortcomings. In the initial stage, the work object is classified using VGG16 with first-person view video, and in the subsequent stage, the action is classified using ST-GCN with skeletal information obtained from third-person view video. This methodological approach facilitates the concurrent analysis of object characteristics and action details, thereby enhancing the accuracy of classification.