Human activity recognition is an important field that increases variety of applications within the field of computer science. Primarily it is classification of human behaviours performed using movement patterns, spatiotemporal signals and their relevant descriptors. Action Recog. is equipped with video data analysis abilities of automatically recognizing and interpreting ongoing operations while showing the subtle nuances of their configuration. These operations may appear in many sizes and shapes, from solitary gestures to exquisitely choreographed social interactions and group actions. This field has numerous applications even in medicine, such as surveillance system, patient monitoring and to facilitate interactions between man and computer. Therefore, accurate recognition of high-level operations is necessary. The presented work considered the remarkable application of HAR for CNN’s permission to categorize human actions automatically. A comprehensive investigation was conducted to address the issue of the possibility of automatic human action identification. It involved a detailed analytical examination and comparison of various network levels to find the best one. The primary criterion for the success of the chosen way is an opportunity to achieve the highest accuracy. The event shows the importance of adequate and reliable activity identification in the conditions of real human interaction and the high level of modern technical progress. Moreover, the presented work not only supports but also theoretically proves the opportunity to use CNN’s to increase the precision of automatic human action identification systems and make them more successful using specially developed neural network facilities.