全部 标题 作者
关键词 摘要


基于计算机视觉的火车钩挡识别
Train Interval Recognition Based on Computer Vision

DOI: 10.12677/HJDM.2015.52003, PP. 17-24

Keywords: 计算机视觉,视频处理,特征提取,特征融合,火车钩档识别
Computer Vision
, Video Processing, Feature Extraction, Feature Fusion, Train Interval Recognition

Full-Text   Cite this paper   Add to My Lib

Abstract:

为了给运煤车喷洒抑尘剂,需要识别火车车厢之间的钩铛。基于计算机视觉的火车钩铛识别只需在站台一侧安装摄像头,对拍摄到的图像序列进行处理从而识别出钩铛。由于光照、天气等诸多因素的影响,首先采用直方图均衡算法对图像进行增强,然后提取图像的LBP纹理特征和直线特征,并进行特征融合,选用SVM (Support Vector Machine)分类器,针对不同的目标通过离线训练的方式获得SVM模型参数。针对四类目标训练了相应的SVM模型,分别为车头、车厢、钩挡和车站,对拍摄的运煤车视频图像可以实现在线识别。经过初步实验验证,提出的钩铛识别系统能够自动识别火车钩铛的开始和结束,从而控制喷洒系统,且安装简单,能够最大程度的减少抑尘剂浪费。
Train intervals between carriages need to be recognized when we sprinkle the dust-restrainer on coal in the train. A train interval recognition method based on computer vision technology is pro-posed, which will set a video camera on one side of the platform to recognize the train intervals by processing image sequences. Because of the variation of illumination and weather, images are first enhanced by histogram equalization algorithm, and then image features including texture features and line features are extracted and integrated. The SVM model will be trained offline for four targets, including locomotive, carriage, interval and platform, and used to recognize the train in-tervals online. Our experimentations show that our system is capable of recognizing the beginning and the ending of train intervals automatically. Furthermore, the proposed system can be embeded in the sprinkle system easily to reduce the waste of dust-restrainer dramatically.

References

[1]  陈春雷, 王阳萍, 祝咏升 (2007) 基于视频列车车号识别系统的研究. 兰州交通大学学报(自然科学版), 6, 99- 102.
[2]  党华丽 (2007) 列车车号自动识别系统研究与设计. 硕士论文, 河北工业大学, 天津.
[3]  邵士媛, 左长进 (2006). 列车车号自动识别系统ATIS应用与实践. 计算机工程与设计, 11, 2108-2111.
[4]  章毓晋 (2006) 图像工程(上册). 清华大学出版社, 北京.
[5]  单勇, 王润生 (2006) 彩色图像混合滤波算法. 数据采集与处理, S1, 67-70.
[6]  邹彬, 潘志斌, 胡森 (2012) 基于局部投影与块LBP特征的图像检索. 中国图象图形学报, 6, 671-677.
[7]  文贡坚, 王润生 (2001) 一种稳健的直线提取算法. 软件学报, 11, 1660-1666..

Full-Text

comments powered by Disqus