In order to quickly and accurately estimate average size of densely packed aggregate particles on a moving conveyor belt, a new image processing method is studied. The method consists of two major algorithms, one is a one-pass boundary detection algorithm that is specially designed for the images of densely packed particles (the word "particle" is used in a wide sense), and the other is average size estimation based on image edge density. The algorithms are cooperative. Our method has been tested experimentally for different kinds of closely packed particle images which are difficult to detect by ordinary edge detections. The new method avoids delineating and measuring every particle on an image, therefore, is suitable for realtime imaging. It is particularly applicable for a densely packed and complicated particle image sequence.