图像识别在编组站驼峰作业过程控制中的应用

Research on the Application of Image Recognition in Hump Yard Process Control at Marshalling Station

利用图像识别理论,针对编组站驼峰作业情况,提出一种通过图像序列快速、直观判断溜放车组特征和走行趋势的方法。根据图像的形状特征,采用伸长度、圆度、矩形度、凹凸度4个参数作为模式样本的特征。采用基于熵的运动目标的检测方法实现目标检测和跟踪。用Kalman滤波来更新背景,用当前帧图像减除动态背景,用信息熵来选择阈值,使目标和背景的信息量达到最大,从而将目标和背景正确分割。通过双轨检测判断车组走行位置。应用统计原理,建立计算距离与实际距离的关系模型,通过分析和计算分别得到双曲线拟合、二阶多项式拟合及三阶多项式拟合三条回归曲线,选取三阶多项式作为关系模型。根据车组的实际位置计算车组的瞬时速度。

By utilizing pattern recognition theory, in allusion to the operation situation of marshalling yard hump, a method is proposed to directly and quickly judge the characteristics and the running trend of rolling train set by means of image sequence. According to the feature of image shape, 4 parameters as elongation,roundness, rectangle and convex concave are taken for sample characteristics. Target inspection and tracking are realized through running target inspection method based on entropy. The background is updated by Kalman filter. Dynamic background is deducted by current frame image. Threshold value is selected by means of information entropy. In this way, the information quantity of the target and the background arrives at maximum, so the target and the background are correctly partitioned. The running location of the train set is judged via double rail inspection. Statistical theory is adopted to set up the relational model of calculated distance and actual distance. By analysis and calculation, 3 regression curves are obtained respectively, namely, hyperbolic curve fitring, second-order polynomial fitting and third-order polynomial fitting. Third-order polynomial is selected as the relational model. Transient speed of train set is calculated according to its actual location.