关键词: 机器视觉, 筒子纱, 定位检测, 显著性检测, 阈值分割

Abstract:

In order to improve the detection accuracy of cheese yarn packages and to reduce wear to the package rods, an improved frequency-tuned salient (FT) significance detection algorithm is proposed to detect the cheese yarn rod position. Firstly, making use of the characteristics of the metal surface reflection of yarn rods head and the farther distance of the yarn rods from the bottom surface, the local illumination of the ring light source was applied to improve the target and background contrast. Then, using a priori knowledge of the target located near the center of the image, a block weighting template was designed to improve the FT algorithm and calculate the image saliency. The maximum between-class variance method (OTSU) automatic threshold segmentation was performed on the salient image to obtain the binary image. Then, the region which was obviously not the target is removed by morphological filtration. Finally, the coordinates of the final yarn rods head position were obtained by Hough transform circular fitting. The field experiment and comparison algorithm show that the improved method has the ability to resist defects. In addition, it has the ability to resist light changes and can be applied to the daytime and nighttime lighting scenes of the factory.

Key words: machine vision, cheese yarn package, position detection, saliency detection, threshold segmentation