A Monocular Vision-based Obstacle Avoidance Android/Linux Middleware for the Visually Impaired
Xiaoyang Qiu, A.J. Keerthi, Teppei Kotake, Aniruddha S. Gokhale · 2019
Increasing maturity in Fog/Edge computing has enabled many latency-sensitive Internet of Things (IoT) applications to achieve better performance and shorter response times. Our work on the URMILA middleware [1] proposed a performance and mobility-aware Fog/Edge resource management solution to support cognitive assistance to the visually impaired. However, URMILA was evaluated only in lab-based emulated scenarios. We overcome this limitation by presenting an affordable, unobtrusive and simple-to-use solution for the visually impaired. Alongside the long cane or the guide dog, our application aims to provide the visually impaired with a more detailed description of their environment. Using a single-camera on the Sony SmartEyeglass SED-E1 as the only sensor and an Android/Linux application, we were able to perform both per-pixel depth prediction and object detection on each image frame. By combining the information from these two sources, we provide users with a descriptive audio feedback assisting them in avoiding obstacles and thus better situational awareness. URMILA is used as before to manage the fog and edge resources in the system. We show the effectiveness of obstacle detection and recognition by creating both an outdoor and indoor scenario.