计算机视觉技术的车型识别应用研究.pdfVIP

一一计嚣．计q 算机视觉技术的车型识别应用研究* 茅正冲。韩毅 (江南大学物联网工程学院轻工过程先进控制教育部重点实验室，无锡214122) 摘要：针对目前立体停车库主要采用传统的传感器入库检测系统来识别车辆信息，存在施工周期长、器件损坏率较高， 克服输入图片背景复杂多变，目标被遮挡的情况，对轿车车型识别这一问题鲁棒性好，具有一定的可行性及应用价值。 关键词：模式识别；车型识别；卷积神经网络；深度学习 中图分类号：TP391 文献标识码：A icationResearchonVehicle Basedon Vision Appl RecognitionComputer Mao Yi Zhengchong，Han ofAdvanced Laboratory ProcessControlfor of of 214122，China) (Key LightIndustry，SchoolIoT，CollegeJiangnan，Wuxi currentstereo thetraditionalsensor to Abstract：The detection thevehicleinformation， garageequipmentmainlyadopts systemidentify butexiststhe suchas construction and problems rate maintenancecost．Soamodelis long period，highdamage high proposed，whichap— vision tOsolvethe ofVehicle Caffeconvolutionneuralnetwork pliescomputertechnology problems recognition．Applying framework， the model and on proceedingfine—tuning basedCaffeNet a excellentmodel optimizationparameteroptimization model，finallyrecognition isachieved．The resultsshowthatthe modelcanhandlethe with andthe obscured． experiment inputimage complexbackgroundtarget Themodelhasa robustnessto car good the ofthe andcertain and value． identifytype feasibilityapplication neural Keywords：patternrecognition；vehiclerecognition；convolutionnetwork；deep learning 引 言