Sistema de visão para detecção de segmentos de planos aplicados em órtese ativa

Abstract

In this work, it is proposed the creation of a computer vision system for an active orthosis in order to guarantee more autonomy movements for its users. Currently, in the market, the main commercialized active orthoses don’t have the means to understand the environment around them, which makes impossible for the orthosis to locate itself in his environment and to predict the user’s actions, for example. In this context, we present a strategy that, using a RGB-D camera, allows to obtain the point cloud of the scenario in order to detect the planar regions of the environment. Assuming that, in urban environments, the vast majority of walking places have plane regions, we can classify transposable regions for the orthosis based on the set of points that can be classified as the floor, for example. In order to be able to quickly detect and classify plans, a series of transformations and filtering is required to reduce the amount of data for processing. Experiments in simulation and practical environments detected the planes of interest and, despite the transformations and filtering performed to ensure faster processing, the shape of the plane remained close to the original one. However, the computational cost for the tested on the embedded computer still demonstrates a bottleneck that must be overcome.