Abstract:

Electric locomotive obtains the power from contact wires along the railway. It is important to guarantee that the train can get the reliable power from the contact wires. Unfortunately, the contact detection approaches can have a huge abrasion on the measurement equipment and have a poor measurement accuracy on the contact wires. This paper, to overcome these limitations, uses a non-contact approach based on stereo vision technology. Support vector regression (SVR) is used to establish the corresponding relationship between spatial points to pixels from four linear array cameras. In details, it trains two SVRs by the pixel coordinates from four linear array cameras to predict the pull-out value and height of contact wires separately. It is the first time that SVR is used to calibrate the linear array camera. Significantly different from the existing methods based on many geometrical relationships, the proposed calibration method avoids plenty of parameters during calibration. The calibration process is quite simple and easy to operate. Besides, the method improves the detection accuracy since it simultaneously combines the information from more than two cameras. In real world applications, it proves that the measurement error meets the actual requirements, and it can avoid the problem that the error grows linearly with the distance increases.

Key words: contact wire detection, linear array camera, calibration technology, support vector regression

摘要：

电力机车通过铁路沿线的接触网获得动力。接触网提供可靠的供电是火车安全运行的重要保证。目前，采用的接触式检测方法对设备的磨损较大。基于三维立体视觉技术，实现了一种非接触的检测方法，首次将回归方法应用于摄像机的标定。该方法使用了4台线阵摄像机，先通过支持向量回归机对采集的图像像素样本进行训练完成标定，建立空间点位置到4台线阵摄像机像素之间对应关系，然后利用得到的模型对接触网进行检测。所提出的线阵摄像机标定方法不同于以往通过几何关系进行标定的方法，避免了标定过程中繁多的参数计算，标定过程简单，操作方便可靠。此标定法能同时融合两台或两台以上摄像机的拍摄结果，提高了检测的准确性。在工程应用中证明了其测量误差满足实际测量要求，且有效解决了误差随距离增大而线性增长的问题。

关键词: 接触网检测, 线阵摄像机, 标定技术, 支持向量回归机