Automatização do reconhecimento de buracos em rodovias usando inteligência computacional

• Ano de defesa: 2020
• Autor(a) principal: Schmidt, Leonardo de Abreu
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Santa Maria; Brasil; Ciência da Computação; UFSM; Programa de Pós-Graduação em Ciência da Computação; Centro de Tecnologia
• Palavras-chave em Português: Redes neurais artificiais; Máquinas de vetores de suporte; Detecçãode buracos; Pavimentação de rodovias; Artificial neural network; Support vector machine; VANETs; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: http://repositorio.ufsm.br/handle/1/22279

Potholes inasphaltpavementsandhighwaysareawell-knownproblem,whichisag- gravatedbythegrowingdevelopmentofautonomousvehicletechnology.Thistechnology, therefore, needstoincorporateefficientautomaticpotholedetectionsystems.Infact,there are differentmethodsforautomaticpotholedetection,however,thesemethodsstillneedtobe further evaluatedbothintermsoftheiraccuracyandspeedofresponseinordertobeusedin real contexts.Inordertocontributewithastepinthedirectionofthisevaluation,thiswork presents theproposalofasolutionforthedetectionofpotholesinimages,beingthedetection of theexistenceofapotholeandthedeterminationofthesameintheimage,theobjectiveof this work.Theworkhastwomainparts,thefirstbeinganimageclassificationarchitecture, which usestheHistogramofOrientedGradients(HOG)techniqueasanextractionofimage characteristics, andtwoclassifiers;ArtificialNeuralNetworksandSupportVectorMachines. The objectiveistooptimizeparametersanddeterminethebestclassifierfortheproblem.The second partpresentsaproposalforarchitecturetodetectpotholesinimages,determiningnot only theexistencebutalsothelocationofapotholeintheimage.Thistaskusestexturede- scriptors from Haralick in asystemof grid applied totheimages,addedtotheclassification architecture ofthefirstpart.Theprocessingtimesofthesearchitectureswerealsoevaluated. The resultspresentedshowthatarchitectureusingtheArtificialNeuralNetworkclassifieristhe best optionforpotholedetection,reaching83% IoU (IntersectionofUnion) in additionto73% accuracyand93%precision.ProcessingtimeisalsofavorabletotheNeuralNetworkclassifier reaching about9 frames per second(FPS).