Design And Implementation Of The Software Of Circular Objects Detection System Based On Blob Analysis

The fast and accurate detection of circular objects has a broad application prospect in the field of machine vision. I made a circular objects detection system combined with the current situation with the current poor-realtime-performance situation, which existed in the process of producing circular hole parts. The system solved the problem of detecting slowly which includes:First, the application:This system took the form of three layers of package to design program. The first layer was the main control interface based on MFC and DirectX. It included loading the local pictures, capturing images by camera, image zoom, displaying image pixel value and RGB component, setting ROI(region of interest), displaying image’s RGB component(used to judge RGB’s sequence), results of processing and so on; The main control interface called the second layer--parameter setting interface--by means of DLL(dynamic link library). This layer included a radius error setting, color setting and so on; The parameter setting interface called the third layer--image processing algorithm--by means of LIB(library). The result of the third layer returned to the second layer, then to the first layer and displayed on the first layer.Second, algorithms:Randomized Hough Transform(RHT) has the advantage of avoiding the huge amount of calculation in the process of transforming one to many mapping compared to the Hough Transform. But the real time of RHT is not up to many customers’requirements in the field of industrial automation. The fundamental shortcoming of RHT is analyzed based on a pixel. This thesis used the following method:making the denoised image to the binary image, finding the candidate circles quickly and accurately by Blob analysis according to a good circle rate which was characteristic of the circle, then finding the real circles by means of the least squares fitting. Compared the both methods, it could be known that the former is analyzed based on a pixel after binarization and edge detection of a image; while the method Ⅰ came up with is analyzed based on pixel groups, which have the same gray value(0or255) after binarization of a image. It could be predicted that the promotion of the velocity of the latter related to the former was very impressive, even if the latter sacrificed the performance of the speed because of the edge extraction then matching circles by means of the least squares fitting(this is to ensure the accuracy of detection of circles). Actually, the detection results also confirmed this prediction.