The work presented is part of a project to locate objects from colour stereo images using hue (or colour). The first step is to convert the images from the RGB camera and frame-grabber format to the HSI (hue, saturation and intensity) format. It is then possible to apply window thresholding to the hue component to extract the silhouette of any target object(s) in the image, knowing the range of hue values which span those of the object(s) to be located. Small areas of the selected hue are removed from the threshold image, leaving the larger areas to be filtered by an edge-detecting filter and then eroded to a 1-pixel wide boundary, defining the outline of the silhouette. A boundary-following algorithm extracts an ordered list of (x,y) coordinate pairs defining the boundary of each sufficiently large area of the selected hue value. The processing of this list of coordinates is the main focus of this paper. Some early results are presented which show promise.
展开▼
