复杂背景下货车制动梁的快速分割方法

Fast segmentation method for brake beam of freight train under complex background

为了准确分割货车制动梁,实现货车运行故障图像动态检测系统(TFDS)的故障自动识别,提出了灰度投影、离散小波变换和自适应分割相结合,利用灰度分布特征与位置相关性实现分割的方法。计算图像的垂直灰度投影,采用离散小波变换分析投影曲线,选择表示制动梁边缘的特征点,拟合其中心线。结合制动梁先验信息,将投影区域旋转相应角度并计算灰度投影,设置自适应阈值选择曲线上的特征波峰,实现不同角度制动梁的分割。实验结果表明,制动梁分割准确率为99%,处理一幅图像的平均时间21.7ms,可以准确地分割不同曝光条件下的图像,为后续的故障检测提供有力的保障。

In order to segment the brake beam of freight train accurately, a segmentation method comprising gray projection, discrete wavelet transform and adaptive segmentation is proposed to realize automatic failure recognition for trouble of moving freight ear detection system （TFDS）. The objects are segmented based on the eorrelation between dis tribution of gray scales and their location. The vertical gray projection is calculated, and the projection curve is analyzed using discrete wavelet transform. The extreme points which represent the edges of break beams are extracted, and the centerline can be fitted accordingly. Considering priori information in the image, a rectangular area in the plane is rotated and gray projection along the long axis is caleulated within the area. An adaptive threshold is set based on the maximum and minimum of projection eurve, and the eharaeteristic peak can be selected. Then, the inclined brake beams are seg mented. Experimental results indicate that 99 percent of the images can be correctly segmented, and average time cost of an image is 21.7 ms. The proposed method can correctly segment break beams in all-weather conditions, even when the original images have poor quality, which advances the engineering application for TFDS.