Selection of Color Contrast Objects Against a Non-Stationary Background Using Modified HSV Model

For the development of such smart systems based on the computer vision as, for example, unmanned aerial vehicles, driverless cars, etc. efficient image processing algorithms are required. These algorithms should provide the possibility to process images in a video stream, perform segmentation, select objects and define informative signatures for further recognition. Objects in optical images usually differ in color, shape, may have shadows, and they stand out against the non-stationary background of the terrain; that is why the HSV color model is often used for their selection. To select objects against the background of the terrain it is necessary to define their signatures and train the recognition system. However the variety of objects and the dependence of their color characteristics on changes in illumination pose a considerable problem. The paper proposes an alternative approach – to train the system to recognize the background, and the fragments of the image that are not recognized as the background should be classified as objects. For real-time estimation of the color parameters of the background it is proposed to convert the traditional HSV model into a one-dimensional signature to speed up calculations during processing. Processing of the images received from an unmanned aerial vehicle by the proposed method makes it possible to select almost all objects of different colors, shapes and sizes against a heterogeneous background and to define their contours for further recognition.