Depth estimation for visually impaired people using mobile devices

Abstract

This paper presents the development of an application to estimate depth using mobile devices that could help visually impaired people to navigate in everyday sites. The mobile device incorporates a stereo camera system and the application takes advantage of the operating system capabilities, such as parallel computation, hardware configuration, in conjunction with the image processing libraries to implement well known and fast performing algorithms for depth estimation. Each camera is calibrated separately and afterwards a stereo calibration is performed. Depth estimation is calculated using two different algorithms and an evaluation of their performance is presented. The mobile devices handle the computational load fast enough to be able to produce the results within the walk time at normal pace. Depth estimation error is low enough to allow object detection like obstacles, room walls or moving persons.