| The average size estimation is an important way to control production quality in many fields, however, the method of traditional measurement by hand is very difficult to meet to the continual mechanized manufacture. A machine vision system based on optics and computers can often be part of a production control loop and provide improvements in manufacturing speed, quality, and compliance with regulations and standards. In recent years, computer vision for industrial inspection has been applied to different applications, such as integrated circuits, steel production, poultry, roads, and crushed rock.At present, most machine detection system obtains average size and shape parameters of particles through segmentation. But to some images with excessive particles, owing to odds of illumination, crude surface, conglutination, overlap, imperfect between particles, the traditional arithmetic of segmentation can not reach very good result. Direct against this kind of situation, the paper presents a method to estimate average size of closely packed particles based on edge density without explicit segmentation. We use, primarily, Canny edges (or edge focused Canny edges) from smoothed images of reduced resolution, and compare with indirect edge density from moment-based thresholding. An average shapefactor is involved in the calculations, obtained for some frames from crude partial segmentation.The results of experiments on a great deal of image sequences of closely packed particles (especially aggregates) indicate that, for closely packed aggregates particles, quite reasonable estimates of number of particles or average size often arise from the method based on the edge density. This method is a relatively computationally inexpensive way of obtaining average size, especially to those image sequences without perfect segmentation result for various kinds of reasons. Thus, edge density based approaches should be possible to use in real-time applications in processing of image sequences.
|
PDF Full Text Request