Abstract:

As trees intrude intoorbits area threaten security of railway transportation, it is necessary to identify and detect invasive trees. Adangerous tree detecting method based on longitudinal parallax is proposed to determine the location relationship between trees and orbits. Multi-scale threshold segmentation is performed on the image using multiple colors. The tree regions are recognized by color feature and fractal dimension. The distances between the tree areas and the center of the view plane are calculated by region matching and moving of two frames. The catenary brackets are detected and measured to optimize the distance estimation of tree areas. Experimental results show that the detection efficiency and measurement accuracy based on longitudinal parallax can meet the needs of real-time detection of railway dangerous trees.

Key words: dangerous tree detecting, tree recognition, image distance measurement, longitudinal parallax

摘要：

侵入轨道区域的树木危害铁路运输安全，需要对入侵树木识别和检测。为判定树木与轨道的位置关系，提出基于纵向视差的危树检测方法。使用多个颜色对图像进行多尺度阈值分割，利用颜色特征和分形维数对树木区域进行识别，再通过前后两帧图像的区域匹配和移动计算树木区域与视平面中心的距离，对接触网支架进行检测和距离计算，优化对树木区域的距离估计。实验结果表明，基于纵向视差的检测计算效率和测量精度可以达到铁路危树实时检测的需要。

关键词: 危树检测, 树木识别, 图像测距, 纵向视差