The article suggests an approach to identifying a potential violator based on images of a person with uninformative distinguishing features on video frames of an intelligent security video surveillance system. The main focus is on evaluating the possibilities of identifying a person in a uniform (recognizing a person by type of clothing) using video stream data and the deeply trained convolutional neural network YOLO. The developed software model makes it possible to increase the likelihood of identifying potential violators by the context of their type of clothing, location, correlation with official duties, and taking into account other factors.