The paper is devoted to the problem of multi-objective control design for visual positioning of moving objects. This research area incorporates the methods of control theory and computer vision, and has a special importance for autonomous vehicles, where visual information from on-board camera gives a rich data about the surrounding world. Visual information can be effectively used in feedback control, for example, in such applications as visual positioning, tracking a visually given line, moving in a changing environment with visual obstacles.

The main objective of this work is to develop the results obtained earlier and design a feedback control algorithm for visual positioning problem based on the multi-objective approach. This approach allows to take into account a set of requirements for closed-loop system performance in different regimes. These regimes, in particular, include the object motion under constant or random external disturbances. The multi-objective structure of the control law has adjustable elements that must be selected in accordance with the imposed requirements. It is convenient to formulate the problems of searching these elements as optimization tasks on the corresponding admissible sets.

The nonlinear mathematical model of object dynamics is considered. In addition, the nonlinear model of the dynamics in image plane of the camera is introduced. The control objective in visual positioning problem is to provide the desired projection of some three-dimensional object of the scene to the image plane. This projection is described by features vector. The main result of the work is the developed feedback control law, which is based on multi-objective structure and computer vision algorithms. Methods for searching adjustable elements of multi-objective structure are proposed. The efficiency of the approach is illustrated by a computational experiment in MATLAB environment.