一种基于改进SIFT算法的轨道板图像匹配方法

摘要

为了提高CRTIII无砟轨道板视觉检测中图像匹配的效率,文章采用一种改进算法完成轨道板图像匹配.传统的SIFT算法使用128维特征描述子,计算欧式距离判断是否匹配,匹配速度慢,并存在误匹配问题.为了提高算法的稳定性和正确率.文中首先对采集图像进行双边滤波,去除噪声并保存图像的边缘信息.在SIFT的基础上采用内切圆选择特征描述子,目的是为了去除远离中心点的那些像素,最后得到了96维特征描述符向量;并且计算皮尔逊相关系数进行评分,通过对相关系数排序,设置适当的阈值,初步找到匹配点,然后运用一种去除误匹配的算法得到最优匹配结果.最终证明,该算法缩短了图像处理时间,并且保证了较高的图像匹配精度和较高的稳定性.%In order to improve the efficiency of image matching in visual inspection of CRTIII ballastless track plate,an improved algorithm is used to complete the matching of track plate images in the paper. The traditional SIFT algorithm uses the 128-dimensional feature descriptor to calculate whether the European dis-tance judgment is matched,the matching speed is slow, and there is a mismatch problem. In order to im-prove the stability and accuracy of the algorithm. In this paper, we first filter the captured image, remove the noise and save the edge information of the image. On the basis of SIFT, we use the inscribed circle to select the feature descriptor. The aim is to remove the pixels away from the center point,and finally get the 96-dimensional feature descriptor vector. We also calculate the Pearson correlation coefficient. By sorting the correlation coefficient,Set the appropriate threshold,find the matching point,and then use a method to remove the wrong match to get the best match results. Finally,the algorithm shortens the image processing time and ensures high image matching accuracy and high stability.