NavSense: A Navigation Tool for Visually Impaired

The visually impaired rely heavily on hearing and touching (with their cane) to navigate through life. These senses cannot make up for the loss of vision when identifying objects in the user's path. In this paper, we propose NavSense, an assistive device that supplements existing technology to improve navigation and peace of mind in day to day life. NavSense relies on range detection, computer vision, and hardware acceleration mechanisms to provide real-time object identification and context to the user through auditory feedback. In particular, we use four hardware platforms - Raspberry Pi 3 B+, Coral Accelerator, Coral Development Board, and Intel Neural Computer Stick - to compare the efficiency of object detection in terms of time and energy during setup and inference phases. Based on these results, it is possible to tailor the design for specific energy-accuracy requirements. Also, we have implemented and used NavSense in real-world scenarios to show its effectiveness.