轨道图像特征点规律分布研究

Research on the distribution of feature points of track images

摘要:

研究轨道图像特征分布规律对提高轨道图像匹配速率，实现轨道三维建模具有重要意义。由于轨道图像背景复杂，图像色彩信息单一，图像特征分布多变，因此对轨道图像特征点分布规律的研究尤为重要。本文首先对轨道图像进行分析，根据轨道图像部件几何特征及布设规律将图形划分为不同区块。运用尺度不变Harris特征点检测算法提取不同区块内图像特征点，Sift描述子对已得特征点进行特征描述并匹配。采用统计方法得出几何特征点的分布规律。试验检测1000幅轨道图像，得到轨道图像几何特征点出现频率分布规律为：挡板座区域（挡板座顶点）97.8%，螺母区域（螺母边角点）57.3%，弹条区域（弹条拐点）53.9%，钢轨轨枕交叉区域（钢轨轨枕交叉边界点）24.7%。

Abstract:

It is a great significance to study the feature distribution law of track image to improve the matching rate of track image and rebuild the three-dimensional modeling of the track. Because the background of the track image is complex, the color information of the image is single, and the features distribution of track image is variable, it is especially important to study the distribution law of the feature points of the track image. In this paper, we analyze the track image firstly. Based on the geometric features and layout of the track components, we divide the image into several different blocks. Then, using the scale-invariant Harris feature point detection algorithm to extract image feature points in different blocks; describe the obtained feature point with Sift descriptors, and finally, match the feature points; Analyzing the geometric feature points distribution of track image with Statistical methods. We experiment 1000 track images which were taken from real word, detect the geometric feature point distribution and got the show up frequency conclusion as following: 97.8% of the baffle seat area (the apex of the baffle seat), 57.3% of the nut area (nut corner point), and the elastic bar area (the elastic bar inflection point) ) 53.9%, rail sleeper intersection area (rail sleeper intersection boundary point) 24.7%.