Tutorial T1A: Assistive Technology for Visually Impaired: Embedded & Vision Solutions

Summary form only given, as follows. The complete presentation was not made available for publication as part of the conference proceedings. Census 2011 classifies more that 5 million people as visually disabled In India. AssisTech (Assistive Technology) group at IIT Delhi aims to develop technological solutions to address their two key challenges; namely independent mobility and access to education. In this tutorial, we will firstly brief about the challenge of mobility and education for visually impaired people. Solutions addressing these challenges could be developed using embedded sensors or with embedded computer vision. Our existing products like SmartCaneTM and OnBoard are embedded system based solutions and are aimed towards facilitating independent mobility. There have been major advances in vision based techniques including possibility of their implementation on low cost embedded platforms. This has encouraged us to visualize a device named MAVI (Mobility Assistant for Visually Impaired). Objective of MAVI is to create a platform that would be able to provide mobility assistance to a visually impaired person in an unstructured infrastructure of developing countries like India. In this tutorial we will discuss specific challenges and various vision techniques used in MAVI to address the requirements of safety, navigation and social inclusion. Designing a complex system like MAVI presents the designer with numerous choices of algorithms as well as platform/hardware resources. We will discuss the challenges in handling system level design complexities and explain our approach towards design space exploration for such systems. Further, we also observe that existing metrics employed in the literature to assess various vision algorithms are inadequate when it comes to complex systems like MAVI. We also propose the need for system level metrics. Finally, we will demonstrate apart from a MAVI prototype, other devices and solutions developed by the AssisTech group.