Obstacle-free paths detection by means of disparity maps for navigation assistance to visually impaired people

Resumen

The detection of surrounding obstacle-free areas is an essential task for many intelligent automotive and robotic applications. It is also of interest in navigation tasks for Electronic Travel Aid devices, which help in mobility and independence to users with physical handicaps, such as blind or partially sighted people. In the last years, one of the most researched topics in intelligent systems has been the obstacle detection, due to the increasing interest in a variety of applications, from autonomous robots to automatic driving systems. Particularly, in the computer vision research area, each year many new references that perform obstacle detection appear, most of them by means of stereo vision or by combining single-cameras with other kind of sensors, such as ultrasound, sonar or lidar systems. This PhD dissertation undertakes the task of developing different methods to solve a related problem, which is the detection of obstacle-free paths in outdoor scenarios by using the information of disparity maps. These disparity maps are obtained by processing stereo correspondences from the left and right images captured by a stereo vision system. Given that the implemented detection algorithms aim to be integrated in a navigation system to support mobility assistance to any kind of visually impaired person, they must fulfil the computational complexity and memory requirements to assure that they are suitable to be processed at real-time frame rates with no-extra dedicated hardware. In the studied methods, fuzzy logic is used to provide certainties of being an obstacle-free path of the analyzed areas, to allow to deal with gradual transitions between obstacles and free paths, and to allow more flexibility and robustness. Experimental results show that the implemented methods provide good results and that they fulfil the system device constraints.