基于机器视觉的摄像整纬纬斜检测研究

Research on Weft Skew Detection of Camera Weft Straightener Based on Machine Vision

摘要:

为了在线检测织物的纬纱倾斜量，研究了基于机器视觉的纬斜检测算法；首先采用改进后的陷波滤波器对织物图像进行频域滤波预处理，得到纬纱较为凸显的纹理图像描述；计算自相关图像初步提取纹理基元，随后构建大尺寸Sobel边缘检测模板，结合Blob分析提取纬纱纹理；再通过霍夫变换将纹理区域变换到参数空间，根据黑塞矩阵分析参数空间图像后，利用阈值逼近手段获取直线检测的阈值，最后反变换到图像空间中，显示出所有检测到的纬斜角；搭建摄像检测的硬件系统和软件界面，检测多类高密、低密织物，检测出的纬斜角度平均绝对误差小于0.35°，平均检测时间在225.5 ms内，能够满足整纬机的实际工作要求。

Abstract:

In order to detect the skew value of fabrics online, a weft skew detection algorithm based on machine vision is studied. Firstly, for the sake of the more prominent weft yarn of the texture , an improved notch filter is used to preprocess the fabric image in the frequency domain. Then the autocorrelation image is calculated to initially extract texture primitives. A large Sobel edge detection template is constructed to extract the weft texture with the Blob analysis. With the texture domain transformed into the parameter space, analyzing the parameter space image with Hessina Matrix is beneficial to obtain the threshold of line detection by the mean of threshold approximation. Finally, the transformation is reversed to display all detected weft skew angles. The camera detection hardware system and the software inference were built to carry out the detection on both high density and low density fabrics. The result showed that the mean absolute deviation was less than 0.35°and the average running time was less than 225.5 ms, which could meet the practical work application required of the weft straightener.