Segmentação e reconhecimento de semáforos com ajuda de instrução especialista

• Ano de defesa: 2018
• Autor(a) principal: Almeida, Thiago da Silva
• Orientador(a): Macedo, Hendrik Teixeira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Programa de Pós-Graduação: Pós-Graduação em Ciência da Computação
• Palavras-chave em Português: Computação; Aprendizado do computador; Aquisição de conhecimento (Sistemas especialistas); Trânsito; Inteligência artificial; Aprendizado de máquina; Conhecimento especialista; Semáforos; Visão computacional
• Palavras-chave em Inglês: Machine learning; Expert knowledge; Traffic light; Computer vision
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: http://ri.ufs.br/jspui/handle/riufs/10771

Traffic light detection and recognition research has grown every year. In addition to this, Machine Learning have been largely used not only in traffic light research but in every field that it is useful and possible to generalize data and automatize a human behavior. Machine Learning algorithms requires a large amount of data to work properly, and because of that, needs a lot of computational power to analyze the data. In this article Expert Knowledge was used in an attempt to reduce the amount of data required by a Machine Learning algorithm. Results show that the inclusion of Expert Knowledge - EK - increased in at least 15% the algorithm test accuracy rate in two different image datasets. The EK used was the traffic light location in a image obtained from a vehicle interior. This idea is based on the theory that there are regions where the traffic lights appear more frequently and, for that reason, those regions have a bigger traffic light appearance probability. Traffic light frequency maps were built to validate this theory. The maps are the result of a human expert analysis over a image group containing traffic images with traffic lights, the human expert tagged in each image the location where the traffic light appeared. The EK inclusion evaluation rates were also superior when testing the detection algorithm followed by the trained classification algorithm, in this test the EK inclusion obtained 83% precision rate and 73% recall rate, while the traditional trained algorithm had 75,3% precision rate and 51,1% recall rate. This article also proposes a traffic light recognition (TLR) device prototype using a smartphone as camera and processing unit that can be used as a driver assistance. To validate this layout prototype, a dataset was built and used to test an algorithm that uses an adaptive background suppression filter (AdaBSF) and Support Vector Machines (SVMs) to detect traffic lights. The application of AdaBSF and subsequent classification with SVM to the dataset achieved 100% precision rate and recall of 65%.