A Control Method Based on Computer Vision and Machine Learning Technologies for Adaptive Systems

Abstract

We consider the problem of organizing the control process in adaptive systems, in which it is required to ensure the preservation of the optimal state of the system when external conditions change. The analysis of existing approaches to its solution showed grea t promise in the synergistic effect of using machine learning and computer vision technologies. A system analysis of the management process using these technologies has been carried out. Its prim ary objects have been formalized, and the research task has been set. To solve it, a method is proposed, the novelty of which lies in the usage of machine learning and computer vision technologies for recognizing and obtaining a compresse d idea of the state of the observed environment, objects of observation and control. And also, the choice of the control team was unified, based on three approaches: a system of rules, a neural network with classification, and machine learning with reinforcement. All stages of the method are formalized, and the possibility of using machine learning technologies (neural networks) for their i mplementation is theoretically substantiated. The practical significance of the developed method lies in the possibility of automating the activities of a human operator in complex adaptive systems through the use of machine learning and computer vision technologies. The method was tested on the example of an adaptive running platform control system. Experimental stu dies have been carried out to assess the efficiency of the method, its perfor mance and accuracy of work in determining the state of objects of observation using computer vision technologies. The result of the work is the proven high efficiency of the proposed approach. The usage of computer vision and machine learning technologies made it pos sible not only to control the adaptive running platform but also to determine critical situations (falling or sudden stop of a person), which increases the safety of the control system, expands its functionality in monitoring the state of the environment and objec ts of observation