| Summary: | Multispectral on-board vision systems of aircraft solve a large number of tasks designed to ensure safe flight in difficult visibility conditions and successful flight performance. These tasks are traditionally divided into low-level and high-level tasks. The low-level tasks – or the primary image processing tasks – deal with the topics of noise reduction in processed images, image enhancements and brightness range detection. Problem-solving methods in on-board computers are subject to severe time restrictions in terms of machine time spent for their implementation. Therefore, the entire set of tasks of both low and high level must be solved in real time. This article presents original algorithms for solving two low-level tasks: discrete Gaussian noise reduction and the edge detection. A modified version of the sigma filter, enhanced by an original algorithm of low computational complexity, was applied to estimate noise level in the processed image and to calculate the optimal cut-off value in the sigma filter. A method, that is, on one hand, analogous to the Canny edge detection algorithm, and, on the other, an alternative one. The differences from the Canny edge detection technique are, firstly, in the use of a vector mask to calculate the estimates of private derivatives in the gradient. This mask provides optimal estimates of private derivatives, in terms of the Least Square Method. Formation of smoothed estimator of private derivatives allowed us to refuse from preliminary smoothing of the processed image in conditions of low intensity noise. Secondly, the proposed method uses a different way of forming edges that provide a choice of "strong" and "weak" lines. Unlike the Canny algorithm, the method used to form and use edges is focused on creating a contour image with a minimum number of short contour lines. Short lines make it difficult to analyze the contour image at the stage of solving high-level tasks. The new edge detection method requires 2-3 times less machine time than the Canny edge detection algorithm.
|
|---|
