Uma abordagem híbrida para planejamento de rotas e controle de navegação de deficientes visuais em ambientes indoor

Abstract

Walking freely in an indoor environment is a huge challenge for visually impaired people due to the lack of information about the scenario. For this audience, there is an abundance of new technologies and navigation models that could be employed for this task, making the relationship between man and the environment as harmonious as possible. However, the navigation of a visually impaired user in an indoor environment is a challenging task, as it requires fast and accurate procedures for locating points of interest and perceiving obstacles, in addition to being offered discreetly and with low acquisition cost. This doctoral thesis aims to specify, build and validate an indoor navigation model capable of allowing a visually impaired user to navigate quickly, safely and with low computational complexity. The methodological approach chosen for this work divides the system into two strategies: a map-based navigation and a non-map based navigation. In navigation based on maps, the construction of address records to compose and the indoor mapping by the Linear Weighted Policy Learner algorithm are made to be used in the elaboration of routes. The representation of the environment has a low density of records per square meter to reduce the complexity and the time to prepare the map. The routing algorithm uses shorter route rules to establish the locations to be visited and increases the coverage of records in the area by applying marker virtualization. The iterative Pedestrian Dead Reckoning (i-PDR) navigation algorithm perceives the reference records and adjusts the user's location in real time, passing instructions through an audio guide. Navigation not based on topological maps uses the stereo vision associated with a musical sound scheme to create a security perimeter around the user to identify the presence of obstacles, giving the sensation of 360 degree sound, covering an angle of 120 degrees to the in front of the user. The results obtained on the error margin and the maximum processing time are 0.10 m and 0.07 s, respectively, and obstacles at ground level and suspended with an accuracy equivalent to 90%.