利用加权有向图的船舶水尺重建与水位识别.pdfVIP

利用加权有向图的船舶水尺重建与水位识别

摘要:针对目前船舶水位识别精度低、检测场景迁移性差等问题,提出一种基于加权有向图的高精度、鲁棒性好

的船舶水尺重建和水位识别方法.首先通过目标检测和图像分割算法识别船体上的水尺字符和水位线;然后根据船

舶水尺的结构特点将识别出来的水尺字符构建加权有向图,通过寻找加权有向图中的最长路径完成船舶的水尺重建;

最后根据重建水尺和水位线的位置关系识别船舶的水位高度.在黄骅港实地采集的视频数据上进行实验的结果表明,

所提方法可以较好地修正前期基于水尺字符检测网络带来的字符误检问题,大大减少因船舶图像背景复杂产生的不

利影响,在10mm的误差范围内,水位识别准确率可达91.3%,显著优于业内主流方法.

关键词:加权有向图;水尺重建;水位识别;字符检测

ReconstructionofShipDraftandWaterLevelIdentificationUsingWeighted

DirectedGraph

Abstract:Inordertosolvetheproblemsoflowaccuracyofwaterlevelrecognitionandpoormigrationof

detectionscene,ahigh-accuracyandrobustmethodofwatergaugereconstructionandwaterlevelrecogni-

tionbasedonweighteddirectedgraphisproposed.Firstly,thewatergaugecharactersandwaterlevellines

ontheshiparerecognizedbytheobjectdetectionandimagesegmentationalgorithm.Secondly,accordingto

thestructuralcharacteristicsoftheship’swatergauge,therecognizedwatergaugecharactersareconstructed

intoaweighteddirectedgraph,andtheship’swatergaugeisreconstructedbyfindingthelongestpathinthe

weighteddirectedgraph.Finally,thewaterlevelheightoftheshipisrecognizedaccordingtothepositional

relationshipbetweenthereconstructedwatergaugeandthewaterlevelline.Thetestiscarriedoutonthe

videodatacollectedonthespotinHuanghuaport.Theresultshowsthatthewaterlevelrecognitionmethod

basedontheweighteddigraphforwatergaugereconstructioncanwellcorrectthecharactererrordetection

problemcausedbythewatergaugecharacterdetectionnetworkintheearlystage,andgreatlyreducethe

adverseimpactcausedbythecomplexbackgroundofthesh