A high cost of equipment that solves the problem of tracking the position and direction of users in real time is one of factors that negatively affect the speed of development of the augmented reality industry. The urgency of this problem is a premise for the development of a financially available tracking system. In this research, we propose a software and hardware architecture of a system that solves three-dimensional tracking problems in a closed space and postures classification using neural network models. Distinctive feature of our system is the feasibility in borders of strictly limited computing power and the absence of any sensors placed on monitored objects. After setting the boundaries of the active area, all the necessary input data is provided by a static camera without an infrared filter. As an example of the implementation of a resource-limited solution, we present the assembly of this solution on a Raspberry Pi version 3 single board computer equipped with the Intel Neural Stick version 2 co-processor and a Raspberry version 2 NoIR camera. The first section of the article describes technical characteristics of the equipment used in the study. The second part is dedicated to the solution algorithm and its brief description. Further, in the third stage, the ways of data collection, necessary for a correct assessment of position, direction and posture are illustrated. The fourth, final section presents the results, discussion and possible directions for further work.