Razvoj hijerarhijske strukture upravljanja mobilnim robotom za praćenje ljudi na bazi robusne stereo robotske vizije

The main topic of this doctoral thesis refers to the development and implementation of the hierarchical control structure in which the algorithms are being executed on the high-level control. By applying the stochastic methods in robotic vision to these algorithms, we can detect people, estimate their position, follow them and recognise their actions in order to carry out tasks where the robot behaves like a human's collaborator. In this thesis, some solutions are offered and present a step forward towards solving the problems that robotic vision system that provides reliable inputs to the control module of the mobile human-collaboration robot, is facing. The robust vision module for human tracking which can be applied in various applications where is necessary for robots to work together with humans and which can be applied on different types of mobile robots, was developed. In this thesis, it is devoted a special attention to the integration, testing and experimental verification of the stochastic algorithms for human tracking such as Kaman and Particle filters, as well as a comparative analysis of algorithms for solving problems of robotic people tracking. A part of the research presented in this thesis is based on a scientific collaboration between the researchers from the Faculty of Mechanical Engineering of Nis and the researchers from the Institute of Automation (IAT) of the University of Bremen. The stereo vision module for a human detection that was developed at the Institute of Automatic Control, University of Bremen (IAT), was used for testing the tracking module which was developed in this thesis. Beside the detection system, the systems for human detection that use 3D sensors, such as the Asus Xtion PRO LIVE 3D sensor were used. The main focus of this thesis was the development of a simulation environment and its control system, as well as the development of modules for human tracking, estimation of a human position and recognition of a human behavior. The simulation environment represents the support to the development and implementation of the real world control system. By adding the appropriate modifications, other mobile robots can easily use this simulation environment. The developed algorithms are evaluated on Faculty of Mechanical Engineering, University of Niš as a part of the doctoral dissertation. In this dissertation, advanced hierarchical control was implemented for the purpose of controlling the mobile robot DaNI, developed by the company National Instruments. This advanced hierarchical control was implemented by using 3D sensor Asus Xtion Pro Live which in the laboratory experimental scenario represents the robotic vision sensor for the detection modules and human tracking. In addition, at the IAT, the vision module which consists of two sub-modules was implemented. These two sub-modules are the stereo vision for a human detection and the tracking module based on the Kalman filter developed in this doctoral thesis