| China is a country which has a wide distribution of rivers and developed water systems.Up to now,China has built 121,000 hydrological stations,which are spread all over the country and are developing rapidly;most of these hydrological stations are located in the Yangtze and Yellow River basins.The water level is an important indicator of navigation safety in the channel,and at the same time,it is also an important factor to ensure the reasonable stowage and safe navigation of ships.As a ruler for measuring water level,water ruler plays an irreplaceable role in water level monitoring.At present,China’s traditional water level measurement method that relies on manual observation is still at a relatively backward level;Although the research and application of artificial intelligence,computer vision and other technologies have made considerable progress in China,and have obtained fruitful research results,however,the research of intelligent visual water level monitoring technology is still in its infancy,and it is even rarer to apply artificial intelligence and computer vision technology to water level monitoring.Therefore,the use of intelligent visual processing technology to improve the existing water level measurement methods and realize the intelligentization of the water level measurement has a wide range of practical application values.This subject relies on the "New Intelligent Water Ruler Automatic Recognition System" project in cooperation with the Yangtze River Yichang Waterway Bureau,and carries out related research work on "the key technology of the new water ruler image segmentation and recognition algorithm".Based on the analysis of the insufficiency of traditional water ruler,this research designed a new type of water ruler for the needs of intelligent visual inspection,and research work was carried out on key issues such as the angle and tilt correction of the water ruler image,the pre-segmentation and precise segmentation of the water ruler image,and the water level recognition in the application.The subject focused on the research of water gauge image segmentation and recognition technology based on the combination of super-pixel and graph cut algorithm,and proposed corresponding solutions.The main research content and specific work of this thesis are embodied in the following aspects:(1)Aiming at the inadequacy of traditional water ruler in intelligent visual monitoring,a new water ruler that meets intelligent visual water level recognition had been designed and implemented;In order to accurately identify the marking of the water ruler,Pre-processing methods such as distortion correction,grayscale processing and grayscale compression and filtering of the water ruler image were proposed;the new water ruler was applied to the recognition of the water level of the Yangtze River,and satisfactory results were obtained.(2)Aiming at the problem of the angle tilt of the water ruler in the acquired water level image,a hierarchical pyramid structure Hough transform tilt correction algorithm is proposed.The algorithm extracts two images from the video sequence of the water ruler to obtain water ruler images of different resolutions,and applies Hough transform to the water ruler images from low to high resolution,gradually narrows the search range of the tilt angle,and finally obtains the water ruler image.The accurate tilt angle of the water ruler,and then perform tilt correction.(3)Aiming at the problem that the water ruler,which has been in the wild environment for a long time,is easily contaminated by stains in the complex natural environment and affects the recognition,a super-pixel pre-segmentation algorithm fused with median filtering is proposed.The algorithm in this paper integrates the idea of median filtering into the selection of the seed points of the super-pixel simple linear iterative clustering(SLIC)algorithm,and changes the clustering center for each pixel in the traditional SLIC algorithm to better filter out stains.The effect of noise on pre-segmentation improves the effect of pre-segmentation.In order to further overcome the influence of water ruler stains and accurately determine the band-shaped area of the water ruler head and measuring part,this paper proposes an improved Graph-Cuts algorithm to realize the precise segmentation of the water ruler.On the basis of pre-segmentation,the algorithm uses the similarity of neighboring pixels to construct the boundary term of the energy function.During segmentation,the super pixel block where the water scale area is set as the target area,and the super pixel block of the non-water scale area is the background area.The improved Graph-cuts algorithm is used for fine segmentation,and the water ruler area is extracted into the barcode area of the head and the spiral stripe area of the measuring part,which is convenient for subsequent identification.(4)The basic framework of water level reading recognition for a new type of water ruler is proposed: the principle of water ruler head area recognition and the principle of expected reading of water ruler measurement.The water level reading recognition is performed on the segmented water level image by the water level recognition principle,and the water level data is obtained.The intelligent visual water level recognition algorithm proposed in this paper is applied to the water level recognition of the Yichang section of the Yangtze River,and the recognized water level data is compared with the real water level data.Both experimental results and actual application results show that the application of the intelligent water level recognition algorithm proposed in this topic has an error of less than 2.5cm in good water quality and an error of less than 3cm in turbid water quality;therefore,the intelligent visual water level recognition algorithm proposed in this topic has good feasibility and effectiveness,as well as broad practical application value. |
PDF Full Text Request