A study on local feature descriptors for point clouds

• Autor(a) principal: Rocha, Luís Cláudio Gouveia
• Data de Publicação: 2017
• Tipo de documento: Trabalho de conclusão de curso
• Idioma: eng
• Título da fonte: Repositório Institucional da UFRN
• Texto Completo: https://repositorio.ufrn.br/handle/123456789/34187
• topic: points clouds; local feature descriptors; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO::METODOLOGIA E TECNICAS DA COMPUTACAO::PROCESSAMENTO GRAFICO (GRAPHICS)

Point clouds are a way of representing 3D data which became very popular due to the rise of low-cost 3D sensors on the market whose output data is represented as a point cloud. Given it low-cost, these sensors have been used used in many different fields, such as games or robotics. In many of these applications, recognizing patterns inside big, unorganized clouds is a fundamental task which is often solved using local feature descriptors, which are a way of encoding information local to a region inside a bigger cloud. Nevertheless, pattern recognition using local feature descriptors is a hard task, whose results nowadays are far from satisfactory (in terms of quality and speed) for most of the non-synthetic scenarios, which motivates the development of new descriptors. As a first series of experiments towards both fast descriptors and descriptors robust to high clutter and occlusion, we develop five descriptors, two of them being simplified (thus faster) versions of existing state-of-the-art techniques, one a totally novel approach to discrete descriptors and two being extensions and adaptations of existing descriptors. Our tests show that although our proposals perform poorly when compared to the state-of-the-art, their simplistic design is enough to achieve reasonable results and perform close to some existing techniques, motivating us to keep improving these results. As a byproduct of our work, we produced a benchmark platform which is open for public usage and improvement, aiming to encourage the standardization of tests with feature descriptors.