基于镜面反射原理进行水位视觉检测的新方法  

The New Methods of Water Level Visual Detecting Based on Mirror Reflection Theory

作　　者：崔振萍[1] 覃永新[1] 叶洪涛[1] 胡波[1] 马兆敏[1] 

机构地区：[1]广西科技大学电气与信息工程学院,柳州545006

出　　处：《科学技术与工程》2014年第15期230-232,236,共4页Science Technology and Engineering

基　　金：教育部科学技术研究重点项目(212135);广西自然科学基金项目(2012GXNSFBA053165)资助

摘　　要：在现有的水位视觉检测技术中,光照和水面倒影的干扰是不可避免的。针对这一问题,利用水面具有镜面反射现象提出一种水位检测的新方法。首先,采集不同水位高度的10幅图像作为训练样本,建立图像像素坐标与相应的实际水位高度之间映射关系的神经网络模型;然后,利用辅助光源照射在岸边产生镜面反射现象,采集水位图像并提取水位高度相应的图像坐标;最后,根据建立的神经网络模型,以200幅水位图像作为测试样本,计算出水位实际高度。在量程为400 mm的精度下,神经网络模型的平均误差为0.02 mm,计算结果的平均误差为0.53 mm。由于减少了镜面反射的影响,取得了较高的精度,为水位视觉自动检测装置的开发提供了理论基础。The interference of light and the mirror reflection on water is inevitable in the existence techniques of water level visual detecting. In order to solve this problem,a new method of water level detection is proposed which utilized the phenomenon of mirror reflection on the water surface. Firstly,capture 10 images of different water level as the training sample,and establish the neural network model of the mapping relations between actual water level and corresponding image pixel coordinates. Then generate mirror reflection phenomenon by using an auxiliary illuminant irradiation on the bank of water,as the same time,capture the water level image and its corresponding image pixel coordinates. At last,as the test sample,200 level images were put into the established neural network model to calculate the actual water level. In the precision of the range of 400 mm,the average error of neural network model is 0. 02 mm; the average error of calculation results is 0. 53 mm. Due to reduce the interference of mirror reflection,a higher precision has been achieved,which provides a theoretical basis for the development of water level automatic vision detection device.

关 键 词：镜面反射 标定 水位 视觉检测 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]