中文摘要：

针对人工判读道岔尖轨伸缩位移图像存在效率低和误差大的问题,为实现尖轨伸缩位移的实时自动监测,提出1种基于层次积分梯度的尖轨伸缩位移图像自动判读算法。采用逐层逼近目标区域的方式,克服尖轨伸缩位移图像中噪声和不相关信息的干扰,以SURF（Speeded Up Robust Features）算子的特征匹配结果为指导,逐步提取图像中的有效区域;利用积分梯度的抗噪特性,根据积分梯度和极值点精确定位刻度尺的特征点位置,结合可信度检验,实现尖轨伸缩位移图像的自动判读。用该算法对监测现场采集的尖轨伸缩位移图像进行判读的结果表明,可在2s内自动判读尖轨伸缩位移图像,总体偏差在0.5mm以内,能够满足目前现场对尖轨伸缩位移实时自动监测的要求。

英文摘要：

Focusing on the problem of poor efficiency and accuracy of the manual interpretation method for the image of switch rail expansion displacement, an automatic interpretation algorithm based on the hierar- chical integral gradient was presented to achieve the real-time automatic monitoring of switch rail expansion displacement. Approaching the target area layer by layer was adopted to overcome the interference of noise and irrelevant information in the image of switch rail expansion displacement. The effective area was extracted step by step from the image under the guidance of the matching results of SURF （Speeded Up Robust Features） operator feature. With the anti-noise characteristics of integral gradient, the feature points of the scale were precisely located according to integral gradient and extreme points. Reliability test was introduced to realize the automatic interpretation for the image of switch rail expansion displacement. The algorithm was applied to interpret the images of switch rail expansion displacement collected from monitoring field. Results show that the proposed algorithm can automatically interpret each image of switch rail expansion displacement in 2 s and the total deviation can be controlled within 0. 5 mm, which can meet the requirements for the real-time automatic monitoring of switch rail expansion displacement.