A Comparison of Three Different Parameterizations to Represent Planes For the Mapping Step of Direct Visual SLAM Approaches

• Filipe Lins IFRN
• Orcilano Luz IFRN
• Adelardo Medeiros IFRN
• Pablo Alsina UFRN

Resumo

In recent years, research in robotics and computer vision has been developed to improve the capacity to construct a more realistic representation of the environment through the use of images. Direct SLAM has been one of the lines of research that have advanced in this direction. The ability to extract information from the entire image allows generating increasingly impressive maps. The use of planes for environmental representation has been the option of several developed techniques. Planes are present in most scenes and can be mathematically represented in various ways. Thus, this article consists of a comparison of three distinct forms of plane parameterization to generate plane estimates using stereo images. For this, an optimization technique used in works on direct SLAM is adopted and experimental results are compared.

Palavras-chave: Simultaneous localization and mapping, Cameras, Visualization, Robots, Jacobian matrices, Sensors, Optimization methods