• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>Proceeding of 2009 international workshop on GIS for transportation >RAIL BLOCKAGE DETECTION WITH SINGLE CAMERA
【24h】

RAIL BLOCKAGE DETECTION WITH SINGLE CAMERA

机译：单摄像头的铁路阻塞检测

获取原文
获取原文并翻译 | 示例
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

In this paper, image processing is proposed to assist rail driver to avoid blockages lying on railroads. With forward-looking single camera mounted on-board, since train normally has high speed, we present a method to initially detect and track rail using templates. To reduce computation further, we could conduct convolution between potential rail points and templates just at some row (for image) intervals. Direction and size of template is the core of this method. Tested in experiments, in the consideration of both speed and precision, proper templates are determined. Once rail points are obtained, generally piecewise-linearis employed to construct the whole rail, though initial rail detection divide rail in image into top and bottom halves and detect them respectively. Considering strong similarity of continuous frames, after initial detection of rail, centred in the former rail coordinates, searching width is narrowed and Kalman filter is employed to improve precision. After checking area has been located, left-right difference of gray average is defined to detect blockage, since compared to no blockage, this parameter changes greatly when blockage emerges. To reduce false alarm, left-right difference of gray average of neighboured segments is computed. If all these results exceed threshold in sequential frames, then blockage indeed exists. Because of relative motion between train and blockage, difference of adjacent frames is used to improve detection precision further. Clustering algorithm is finally adopted to get the final blockage.
机译：在本文中，提出了图像处理以帮助铁路驾驶员避免铁路上的障碍物。通过将前瞻性单摄像头安装在车上，由于火车通常具有较高的速度，因此，我们提出了一种使用模板来初始检测和跟踪铁路的方法。为了进一步减少计算量，我们可以仅以某一行（对于图像）间隔在潜在轨点和模板之间进行卷积。模板的方向和大小是此方法的核心。经过实验测试，同时考虑了速度和精度，确定了合适的模板。一旦获得了轨道点，通常采用分段线性的方法来构建整个轨道，尽管最初的轨道检测会将图像中的轨道分为上半部分和下半部分并分别进行检测。考虑到连续帧的强相似性，在对铁轨进行初始检测之后，以前一个铁轨坐标为中心，缩小搜索宽度，并使用卡尔曼滤波器提高精度。在确定检查区域之后，定义灰度平均值的左右差以检测阻塞，因为与没有阻塞相比，此参数在出现阻塞时会发生很大变化。为了减少误报，计算相邻段的灰度平均值的左右差。如果所有这些结果在连续帧中都超过阈值，则确实存在阻塞。由于列车与障碍物之间的相对运动，相邻帧之间的差异被用来进一步提高检测精度。最后采用聚类算法得到最终的阻塞。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号