Figure 3-5 Detection Flowchart
The key technical points are as follows:
a) In order to minimize noise and other factors in the image, the image needs to be preprocessed (Figure 3 -6). First, median filtering is performed on the image to smooth the noise. Secondly, the image is threshold-divided to roughly segment the target region. Then, the image is closed to further eliminate the noise of the target region. After the above processing, the noise in the target area has been well suppressed. The target area is then accurately positioned. To this end, the processed image is scanned from the inside out to obtain an inner (outer) edge point set, and then the edge point set is elliptically fitted.
Figure 3-6 Bottle machine vision inspection systems manufacturers
b) In order to facilitate the direct traversal of the defects on the ring, the polar part of the ring is transformed (Fig. 3-7), and the circular window (inspection area) is converted into a square, by comparing the circumference and the radial direction. The upper concentration is used to detect scars.
Figure 3-7 Polar Transformation
c) Use the LSD algorithm to detect the straight edge of the preform body and accurately and quickly position it into the bottle area (Figure 3-8). The detection result of sub-pixel precision can be obtained in linear time. It can be applied to any digital image without debugging parameters, and can control the number of false detections by itself. LSD is an algorithm for locally extracting lines, which is faster than Hough.
This article is from https://www.topvision.net/ .