Users utilize an accessible smartphone app to identify and orient the ballot paper.
Through provided coordinates, the correct position on the voting document can be located.
Small stencils are placed at the precise coordinates to cast a vote.
Building an AI-Powered Accessible Prototype
The smartphone application is the central element of the entire process. As a prototype developed during our Bachelor's thesis, we created an accessible web application that allowed us to conduct extensive user testing.
The app serves two key functions: First, it identifies and correctly orients the voting materials. Second, it guides users through the complete process with step-by-step instructions and provides the correct coordinates for each ballot.
For document recognition and orientation in the prototype, we experimented with various technologies. We trained our own AI using TensorFlow and Google's Teachable Machine, and utilized Optical Character Recognition with Tesseract.js.
The prototype had to be fully tailored to the needs of blind and visually impaired individuals. Accordingly, it is barrier-free, fully compatible with VoiceOver, and features a simple, high-contrast UI design.
Co-Designing Solutions with Affected Users
Since we ourselves are not affected by visual impairments, it was crucial to establish the best possible dialogue with affected individuals. Beyond numerous conversations and testing sessions, a series of co-design workshops formed the central element of our approach, where we explored how such workshops can be conducted accessibly.
We conducted four major workshops with a small working group of affected individuals. Throughout these workshops, we completed a full design process from problem definition through to solution development.
Since design workshops typically rely on sketching to present and explain ideas, alternatives were necessary. We used clay and Lego together with workshop participants to visualize ideas.
Another challenge was substituting the commonly used Post-its. We developed a board matrix where pins with different haptic properties could be used to plot topics across a two-dimensional space. This allowed us, for example, to categorize concerns on a scale of relevance versus feasibility.