模具监视系统中的图像快速配准算法

An Efficient Image Registration Method in Mold Monitoring System

注塑机的振动及模具归位不准等原因容易造成目标图像与样本图像间的偏移,从而引起模具监视系统产生误检和误报。为此,提出一种基于尺度不变特征变换描述子的实时图像配准算法。该算法采用一种快速角点检测算法提取图像关键点,围绕关键点构建128维描述子,依据最近邻匹配法得到特征匹配对,最后通过随机采样一致性(RANSAC)和最小二乘法剔除误配点并拟合变换参数。最后,用实验证明了该算法的有效性。

Considering the fact that mechanical vibration and mold homing inaccuracy can result in an offset between target image' s and template image and can cause false detection and alarm of mold monitoring system, an efficient image registration method based on scale invariant feature transform(SIFT) was proposed to extract image key points via a fast corner detector and to build 128-D descriptors around these key points,and to get corresponding matching points based on the nearest neighbor method,as well as to obtain the mapping relationship between images by using RANSAC and least squares techniques.Experimental results prove the effectiveness of this method.