Densidade de ocupação espacial como representação para comparação de nuvens de pontos e aplicações

Abstract

In this work, spatial density maps are proposed as a high-level representation for objects given by point clouds in applications that require comparison between different objects. Traditional approaches present complex metrics, which includes statistical information associated with the underlying point distribution. They are highly dependent on the model and on the number of distributions of the set. Our method obtains a representation by spatial density maps in linear time, allowing query in constant time and efficient comparison between different sets. The spatial density pattern is constructed such as to obtain an implicit volume for the data or saturated histogram of points per voxel. The advantages of this representation are exploited in the problems of detecting changes in 3D environments for robot navigation and recognition of human actions in depth map sequences. In both cases, the spatial density maps representation achieved superior results compared to state-of-the-art methods.