The article presents a hybrid approach to recognizing human actions, combining neural network extraction of skeletal features with deterministic geometric analysis based on vector algebra and affine transformations. A review of research on this issue has been conducted. Unlike traditional solutions that require re-training the model when adding a new action, the proposed system allows the user to dynamically set and modify a set of recognizable actions without the involvement of a specialist in the field of machine learning. Each action is defined as a sequence of poses described by the relative location of key body points. The comparison of the current and reference poses is carried out through the cosine similarity of the vectors, and resistance to angle changes is provided by three-dimensional affine transformations. The software is implemented in Python using the MediaPipe and OpenCV frameworks, has an intuitive graphical interface, and works with a regular webcam. Experimental testing has confirmed the correctness of recognition of specified actions with an accuracy of at least 85% under natural conditions. The solution is focused on application in behavior management systems in organizational environments where flexibility of configuration, interpretability and a low entry threshold are important.

Keywords: human action recognition, vector algebra, affine transformations, hybrid model, behavioral management, human–machine interfaces