Identificação de pedestres por meio de mapas de densidade construídos com ASIFT e fluxo óptico

• Ano de defesa: 2016
• Autor(a) principal: Lucchetti, Marcus Vinícius Fitz
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Espírito Santo; BR; Mestrado em Engenharia Elétrica; Centro Tecnológico; UFES; Programa de Pós-Graduação em Engenharia Elétrica
• Palavras-chave em Português: Optical Flow; Density Map; Pedestrian Detection; Imaging Processing; ASIFT; Binary Hierarchical Clustering; Detecção de pedestres; Clusterização hierárquica binária; Processamento de imagens; Visão por computador; Sistemas de reconhecimento de padrões; Aprendizado do computador; Vigilância eletrônica; Fluxo óptico; Engenharia elétrica; 621.3
• Link de acesso: http://repositorio.ufes.br/handle/10/9557

Pedestrian detection in digital images is a relevant paradigm in areas such as security and entities tracking. Recent worldwide events involving investigation of suspicious actions in public places, such as airports, subway stations and crowd manifestations justify the direction of researches on this subject. Different clothing styles, colors, brightness, articulations, intra-inter people occlusions and similarities among boundaries of objects are some of the obstacles to be overcome by researchers. Under this perspective, this work contributes with a proposal to detect people in crowd. The presented method combines a people detection technique with a density map generated by relevant keypoints detected using ASIFT and validated with optical flow. The selected keypoints still allow an additional step of pedestrian prediction, which is performed frame by frame from binary clustering among the possible location of a pedestrian and the detector output. This research still purposes to generalize the density map thresholds in order to make the integration with pedestrian detectors that return some type of score. The experimental results indicate that this approach is more accurate than traditional methods.