Research On The Segmentation Algorithm In High-Speed And Real-Time Inspection Of Complex Curved Surface Defect

In the defect detection of complex curved surface, surface aberration of the image is very large, magnification at different points is also unequal, and surface discrete degree of light field intensity is very large. For these reasons, it is very difficult to segment the image of complex curved surface. Therefore, it is also difficult to meet the practical requirements of high detection speed and precision, even if the adaptive local threshold segmentation method based on integral image whose function is strong is adopted. As the improvement of the adaptive local threshold segmentation method based on integral image, an algorithm of automatic selecting threshold is proposed. Firstly, the initial threshold is set. Secondly, false detection rate is counted. If requirement is not satisfied, the threshold is adjusted according to the dichotomy, until the requirement of false detection rate is satisfied. The threshold above is the optimal threshold, and they are applied to the adaptive local threshold segmentation. Experimental results show that this method can greatly improve the detection accuracy and speed of complex curved surface in the condition of uneven illumination, taking bullet surface inspection as an example, accuracy rate is more than 99%, segmentation time is less than 300 milliseconds, and the industrial requirements of real-time inspection has been satisfied.