论文部分内容阅读
随着我国高速铁路通车里程不断增加,高速铁路的运营安全备受关注,异物侵入铁路限界对运营安全危害极大,有效检测侵入线路净空的异物对保障高速铁路安全运营具有重要意义。铁路场景环境光线多变和图像通道众多的特点对基于图像的异物检测方法的处理效果和实时性提出了较高的要求。针对铁路场景抖动发生在垂直方向的特点,提出了一维灰度投影结合高斯滤波的图像快速去抖方法,在大幅提高处理速度的同时获得了较好的去抖效果;针对复杂多变的背景,提出了一种基于前景目标统计分布的背景更新算法,定义了目标分散指数用于确定行列投影次序,通过统计前景目标分布实现背景更新,在提高速度的同时解决了传统背景更新算法难以解决的鬼影问题。最后通过背景差分获取前景目标,并通过目标标记、合并与特性分析提高目标检测的准确性。沪宁城际高速铁路典型场景的现场实验表明,该算法能有效检出铁路场景侵限目标,系统综合误检率约为0.54%,漏检率为0。
With the increasing mileage of high-speed railways in our country, the operational safety of high-speed railways has drawn much attention. The entry of foreign bodies into railroads is extremely detrimental to operational safety. It is of great significance to effectively detect foreign bodies intruded into the airways to ensure the safe operation of high-speed railways. The change of ambient light and many image channels in railway scene put forward higher requirements on the processing effect and real-time performance of the image-based foreign object detection method. Aiming at the characteristics that the railway scene jitter occurs in the vertical direction, an image fast de-sharpening method based on one-dimensional gray projection combined with Gaussian filter is proposed, which achieves a good debounce effect while greatly increasing the processing speed. In view of the complex and ever-changing background , This paper proposes a background updating algorithm based on the statistical distribution of foreground objects. It defines the target scatter index to determine the projection order of rows and columns, and updates the background by statistical foreground object distribution. While improving the speed, it also solves the problems that the traditional background updating algorithm is difficult to solve Ghost problem. Finally, foreground targets are obtained by background difference, and the accuracy of target detection is improved by target labeling, merging and characteristic analysis. The field experiments on the typical scene of the Shanghai-Nanjing intercity high-speed railway show that the proposed algorithm can effectively detect the intruding target of the railway scene. The system comprehensive false detection rate is about 0.54% and the missed detection rate is zero.