The development of effective methods for recognizing human actions is an urgent task from both a scientific and an applied point of view. In recent years, research has focused on Human Action Recognition (HAR) in video analytics. Convolutional neural networks (CNNs) are widely used for image classification tasks. Deep learning stands out as a more differentiated and effective method of recognizing human actions. Among the methods, YOLO is one of the most frequently used. The advantage of the YOLO algorithm over other algorithms is that it works in real time, activating the neural network only once when motion is detected in the image. The article considers application of the YOLO algorithm for classifying HAR tasks. The issue of recognizing several different dynamic human movements (standing, sitting, running and walking) was considered. The results are implemented in the MATLAB software package.