Image sampling techniques and requirements for automated image analysis of rock fragmentation

Abstract

When photographing or video taping rock fragmentation, sampling strategies have to be carefully considered, in order to produce an image that is both capable of being analyzed, and representative of the entire rock assemblage. A consideration is where the image is to be taken. It could be the top of a muckpile, the front edge or a cross section of the muc k pile as it is being mucked. It could be photographed in the back of a haulage truck, the bucket of a loader, or on a conveyor belt. Or it could be photographed in a stockpile. All strategies are valid, but each must be evaluated to ensure good picture quality (i.e. not obscured by dust), and representative sampling (i.e. not skewed because of some size sorting by mechanical processes). Another consideration is the scale of the image. At the very least, the area must be limited so that the individual blocks can be recognized by the image analysis algorithms. If zoom-merging techniques (images at different scales) are to be used, an appropriate strategy must be employed so that the relative number of images at each scale somewhat reflect the differences in scale. Furthermore, when zooming, care must be taken to avoid pointing the camera at zones of large or small blocks that might attract the eye. A final consideration is the angle of the surface being photographed with respect to the camera. Ideally the surface should be perpendicular to the direction the camera is pointed, as that eliminates perspective error. Perspective error can also be reduced by using camera lenses with long focal lengths where possible. As an alternative, the image can be rotated (tilted) to compensate for oblique angles. This requires two perpendicular scale bars to be appropriately placed in the image.