Técnicas de remoção de fundo em imagens estéreo obtidas a partir de câmeras em movimento

Abstract

Usually, during the acquisition of one or more images, several structures are recorded beyond the object or region of interest. A computer vision application may have its specific focus, but the images obtained will include, in addition to this focus, several other objects or textures that are in the same scene. Assuming that the object or region of interest is in the image foreground, to segment the pixels of this image between those belonging to the background and those really belonging to the foreground becomes a fundamental task for any computer vision system. In this work, the segmentation approach of the image is analyzed using the concepts of stereo vision, i.e. obtaining images from two (or more) cameras and, after processing them, identify the pixels belonging to the objects more distant from the cameras and consequently belonging to the background of the scene. In addition to the mathematical concepts of stereo vision, details are presented regarding image preprocessing, local descriptors, sparse matching and other algorithms required for this task.