Estimação de densidade de multidões: proposta de arquitetura

• Autor(a) principal: Silva, Weryques Santos da
• Data de Publicação: 2021
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFG
• Texto Completo: http://repositorio.bc.ufg.br/tede/handle/tede/11510
• topic: Visão computacional; Reconhecimento de objetos; Estimação de densidade; Viés de perspectiva; Homografia; Homography; Computer vision; Object recognition; Density estimation; Perspective dias; CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::SISTEMAS DE COMPUTACAO

The estimation of crowd density depends directly on the accuracy of the location of peoplein the environment. Current work estimates density directly on security camera viewsand uses a diagonal perspective of the environment. The diagonal perspective causesthe perception of the distribution of the crowd of people to be non-uniform and withdistortions in the real distance between the people located in the image, the PerspectiveBias problem. To deal with this problem, a system was built, the HMODE, with anarchitecture based on perspective transformation (homography) for density estimation.Homography provided greater reliability for the generated density map so that the densityis now estimated on the transformed coordinates for a top view. The system wentthrough two evaluation cases. In the first case, a new dataset was generated from multiplecamera views. The second case of evaluation used an existing dataset, PETS 2009. Theevaluations made it possible to evidence the feasibility and adaptability of the HMODEarchitecture to the environments.