| 摘要 |
<p>The invention relates to a method for the automatic segmentation of coloured images in order to determine the prehension points in uniformly coloured objects, intended to be applied in redundant articulated robots provided with prehension elements. According to the invention, the method comprises a first stage of acquisition (1) of a coloured image, followed, in a second stage, by the image segmentation (2) using the HSI colour model (Hue, Saturation, Intensity). By the segmentation of the hue and saturation planes which contain the colour information, the object of interest is separated from the image background. The two resulting binary images are then merged by applying a logical AND operation. In order to obtain an optimal segmentation, i.e. an object separated from the background as well as possible, the parameters of the segmentation operation are automatically adapted by using a control algorithm based on an extreme point search algorithm. The controlled variable (I), also called uncertain quantity, represents a quantization of connectivity of object-pixels in the segmented image. The optimum of the segmentation operation is given by the minimal value of the uncertain quantity (I). The resulting binary image is supplied to the attribute extraction block (3) and classification block (4) meant to determine the key points of the viewed object. In the end, these key points are employed in the calculation (5) of the prehension points of the object.</p> |
