【作者】 郭保青；

【导师】 唐涛；

【作者基本信息】 北京交通大学 ， 交通信息工程及控制， 2008， 博士

【摘要】 综合检测列车是重要的基础设施检测装备,是高速铁路安全运营的保障。实时高精度的位置信息是综合检测列车准确检测的前提,研究一种高精度自主定位方法对高速综合检测列车国产化和提高我国既有线基础设施检测水平具有重要的理论和现实意义。论文分析了综合检测列车定位需求的特点,首次提出视觉应答器(VisualBalise,VsB)的概念,探讨了利用地图匹配和图像特征识别轨旁及轨道自身视觉应答器,并用它们校正里程计累积误差从而实现综合检测列车连续、高精度、车载自主定位的方法。该方法包括确定视觉应答器捕获区间,视觉应答器图像特征识别与定位,视觉应答器实际精确位置计算三个步骤。为了确定列车是否进入视觉应答器的捕获区,需要进行列车位置估计。论文首先提出了基于GPS与稀疏栅格地图匹配的列车初始定位方法,随后结合GPS位置输出和线路先验数学模型,利用最近点估计和极大后验概率估计方法给出了直线轨道的列车位置估计;针对上述两种估计方法在曲线上计算量大和曲率匹配无法在圆曲线上定位的弊端,提出了基于曲率和航向组合匹配的曲线列车定位方法。针对轨旁视觉应答器的识别,提出了利用特征点虚拟轨迹确定搜索区域,利用轨迹线灰度投影将轨旁视觉应答器从图像中分割出来的方法;并利用支撑矢量机对分割出来的图像进行了分类和识别。针对线路自身视觉应答器,首先将道岔特征检测转换为图像中直线的检测,提出了利用图像截面灰度投影生成候选轨道集合,利用Hough变换验证轨道真实性的检测方法。该方法克服了标准Hough变换计算量大的缺点,同时利用了Hough变换抗噪性好的优点来验证轨道真实性,因此能够快速准确地检测出轨道特征。为计算视觉应答器的实际位置,论文提出了基于单幅图像的一维简化定位模型,分析了模型误差;论文还利用平行钢轨轨距为常数的特征,提出了等比率立体三角定位方法,相对于平面三角定位提高了测量精度。理论计算和现场实验均表明,利用图像特征识别视觉应答器,并将其校正里程计误差的定位方法能够满足综合检测列车的定位需求,为综合检测列车提供了一种高精度低成本的定位解决方案。更多还原

【Abstract】 Comprehensive monitoring train is an important infrastructure detecting facility which ensures normal operation of highspeed railway.Accurate position is the basis of precise detection.A research on autonomous train location method is of great theoretical and practical significance for localization of comprehensive monitoring train and enhancing the infrastructure detecting level of existing line.This dissertation analyses position requirement of comprehensive monitoring train. It puts forward the concept of Visual Balise(VsB) for the first time.A novel positioning method to identify trackside and track-own VsB with map matching and image recognition is proposed.The identified VsBs with accurate position are used to calibrate odometer accumulative error.There are 3 steps of this method:determination of VsB capture area;recognition and location of VsB image features,and VsB’s actual precise position calculation.To determine the VsB capture area,the train rough position should be estimated. Firstly,A sparse grid map matching method is presented for initial train positioning. Then,the nearest point estimation and maximum posterior probability estimation are used for estimating train position based on GPS output and track’s mathematics model. And a position estimation method based on curvature and course combined matching is also proposed in curve tracks.As for trackside VsB,a mechanism for restricting searching area with virtual track is introduced to improve searching speed.In the restricted area,grey projection is used for partitioning VsB features out of whole image.Finally,the partitioned VsBs are classified and identified by SVM.As for track-own VsB,the frog of turnout is firstly abstracted as intersection of straight tracks.Then a straight track candidate pool is generated by grey projection of image line scan,and Hough transform is used to verify the authenticity of candidate tracks.Since the defects of heavy calculation load for standard Hough transform is overcomed and the anti-noise advantage of Hough transform is used to verify the authenticity of tracks,this method can quickly and accurately detect the straight tracks in images.For calculating actual precise position of VsB,a simple 1-D calculation model based on single image is used to reduce calculation load.Theoretic error arising from height and lean angle of camera is discussed.Another 3-D triangle positioning method with gauge constant is also discussed.Both theoretical calculations and experimental results show that calibration of odometer error with Visual Balise(VsB) identified by map matching and image recognition can meet train positioning requirement.And it provides a high-precision and low-cost solutions for precise autonomous train location.更多还原