改进霍夫变换与像素分割的水位识别方法

Improved water level recognition by Hough transform with pixel segmentation

基于图像的前端水位测量平台可以显著提升水位监测系统的效率。由于一年中水流情况各不相同,大多数摄像仪器并不能保证测量的稳定性。为防范化解超标准洪水信息采集可能带来的风险,本文提出了一种改进霍夫变换与像素分割的水位识别方法,将霍夫变换应用于不同分辨率的水尺图像,以对倾斜水尺目标进行校正。利用中值滤波解决水尺污渍的读数问题,在预分割的同时过滤了修正后的水尺图像。最后将相邻像素相似度函数的边界项纳入图像分割算法,提升水尺和水面区域的分类精度以获得准确的水位数据。实验结果表明,改进后的水位识别算法能够在极端恶劣的气象条件下,提供50ms/次的水位识别,识别误差小于2cm,确保了在超标准洪水过程中出现缺乏光照、水面阴影、水尺反光等情况时,水位识别监控的准确性和可靠性。

An image-based front-end water level measurement platform can significantly improve the efficiency of water level monitoring systems.As the water flow varies throughout the year,most camera instruments do not guarantee the measurement stability,in order to prevent and resolve the risks that may be brought by super-standard flood information collection.We propose a water level recognition method with improved Hough transform and pixel segmentation,and apply the Hough transform to water scale images with different resolutions to correct for skewed water scale targets.The median filter is used to solve the reading problem of water scale stains,and the corrected water scale image is filtered while pre-segmentation.Finally,the boundary term of the neighboring pixel similarity function is incorporated into the image segmentation algorithm to improve the classification accuracy of the water scale and water surface area to obtain accurate water level data.The experimental results show that the improved water level recognition algorithm can provide recognition error less than 2cm and recognition speed up to 50ms under extreme bad weather conditions,ensuring the accuracy and reliability of water level recognition monitoring in flood season under the influence of extreme weather such as lack of light,water surface shadow and water ruler reflection in super standard flood.