Introduction. Methods for processing information contained in variations in the optical flow intensity during object motion are widely used in numerous technical applications, including space research, technical diagnostics, machine (technical) vision, object tracking in digital images, autonomous navigation of unmanned vehicles, etc. Among these methods, monocular techniques for estimating the parameters of the video camera own motion have demonstrated the greatest practical efficiency, both based on a cartographic analysis of the underlying surface and using various algorithms for estimating optical flow (velocity field) parameters. Although existing velocity field estimation algorithms offer advantages such as operability in the absence of terrain maps and computational costs easily implemented in onboard processors, their practical application is significantly complicated by the inevitable noise in optical measurements, which can have a wide variety of physical origins and reduce dramatically the accuracy of optical flow parameter estimation. Therefore, the objective of the research discussed in this paper is to address the problem of simultaneous high-precision estimation of the intensity of optical flow and identification of its parameters under conditions of measurement noise with unknown probabilistic characteristics. A theoretical solution to this problem enables the development of a new general approach to the synthesis of robust algorithms for high-precision optical flow processing in video monitoring systems. This, when applied in practice, will ensure noise immunity and the required accuracy characteristics for machine vision systems, autonomous navigation systems of unmanned vehicles, and other applications.

Materials and Methods. The solution was obtained by monocular methods for determining the camera proper motion and minimizing a regularized quadratic criterion. The starting-point for the solution was the formulation of the problem as one of stochastic estimation and parametric identification of a discrete linear nonstationary system while observing its state vector under interference with an unknown probability distribution. The synthesis of the “estimation-identification” algorithm in this formulation was implemented as a procedure that guaranteed the highest accuracy in the minimax sense. Minimizing the resulting minimax criterion allowed us to construct an algorithm for estimating and parametrically identifying the optical flow as a stable vector recurrent procedure, easily implemented in onboard computers of moving objects.

Results. The problem of simultaneous high-precision estimation of the optical flow intensity and identification of its parameters under measurement noise with unknown probabilistic characteristics is solved. This problem has not been considered in the scientific literature to date. The solution will enable the development of an approach to the synthesis of robust algorithms for high-precision processing of optical flows in video monitoring system. In practical use, it will ensure noise immunity and the required accuracy characteristics of machine vision systems, autonomous navigation systems of unmanned vehicles, etc. The assessment of the practical applicability of the developed algorithm for estimating and identifying optical flow parameters was performed under conditions of non-Gaussian measurement noise through numerical simulation. Despite the specified high level of measurement noise, the errors in estimating the optical flow intensity at all tested coordinate points proved to be both rapidly converging to steady-state values and very small in the steady-state mode (a few percent of the maximum value of the optical flow intensity).

Discussion. The obtained data confirm that the proposed algorithm has such advantages over known optical flow processing algorithms as the ability to estimate the intensity of the optical flow and identify its parameters under noise conditions, whose probability distributions are a priori unknown. It is characterized by high accuracy and robustness, and it does not require high computational costs.

Conclusion. The practical significance of the developed algorithm consists, firstly, in the possibility of high-precision stable processing of the optical flow under the conditions of uncertain probabilistic nature of measurement errors, and secondly, in the computational efficiency of the developed “estimation-identification” procedure. This, in turn, provides its successful practical application in solving optical information processing problems in vision systems, navigation of autonomous robotic systems, space exploration, technical diagnostics, and other fields.