An information delivery system for visually impaired people in dynamic environment

In this paper, we propose an information delivery system for visually impaired people to examine any object in their environment. This system provides a magnified view of any region that is pointed at by the user to monitor it. The human pointing gesture has the ability to extract an element from the environment. The two important problems in codifying this gesture are to estimate the direction to which the finger is pointing to, and then to segment the region of interest according to this direction. To solve these two problems, we present a hand and a finger detection method using color and disparity data from a stereo camera system. Then, we estimate the direction in 3-D using geometric and textural features. We utilize the estimated direction to discriminate potential candidates in the 3-D space. Finally, we segment the region of interest by backprojecting the candidate areas while filtering signal noises of the disparity data. Experimental results that validate the proposed system in a dynamic environment is also presented.