I-VAMOS: Independent Voting with Accessible Multimodal Offline System for Visually Impaired Users
Gyeongdeok Kim, Chungman Lim, Gyungmin Jin, Gunhyuk Park · 2026 · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI '26) · doi:10.1145/3772318.3791336
Summary
This paper introduces I-VAMOS, an offline multimodal voting assistance system designed to enable blind and low-vision (BLV) voters to cast paper ballots independently and secretly. The authors situate their work in the constitutional right to a secret ballot, noting that BLV voters routinely lose that right in practice: existing tactile overlays and braille sleeves require braille literacy (which is declining) and often still demand assistance from poll workers or companions, while electronic ballot-marking devices (BMDs) raise costs, security concerns, and have even been ruled unconstitutional in Germany and withdrawn from use in the Netherlands. Online voting is dismissed as sacrificing ballot secrecy and verifiability for accessibility. Building on three participatory design sessions with a low-vision co-designer plus a follow-up workshop with three additional BLV participants, the team iteratively refined a system combining: (1) a ballot slide frame that constrains the ballot to move only vertically, (2) a fixed spring-loaded stamp eliminating fine-motor alignment, (3) a webcam plus Intel N100 mini-PC running Tesseract OCR, (4) an Auditory Feedback Manager that speaks candidate number, party, and name with proximity cues, and (5) a Visual Enhancer that magnifies, thickens strokes, and applies WCAG 2.1 SC 1.4.6-compliant high-contrast color pairs. The entire system operates offline and costs a few hundred USD, compared with US$3,000-8,000 per BMD unit. The authors then conduct a within-subjects user study with 16 BLV participants balanced across vision status, braille literacy, and age, comparing I-VAMOS against official Korea Election Commission (KEC) tools (tactile sleeve and magnifier).
Key findings
I-VAMOS significantly outperformed the KEC tools on nearly every measure. Stamping accuracy rose from 75.0% with KEC tools to 91.7% with I-VAMOS, with OCR precision of 98.9%. System Usability Scale scores rose from 52.8 to 79.1, and NASA-TLX workload dropped from 46.2 to 26.1, with significant reductions in mental, physical, temporal, and frustration subscales. Crucially, I-VAMOS also narrowed disparities across subgroups: accuracy for blind participants jumped from 66.7% to 91.7%, braille-illiterate users went from 66.7% to 91.7%, and older voters improved from 62.5% to 87.5%. The one trade-off was completion time: tasks took 29.6s with I-VAMOS versus 18.2s with KEC tools, due to added OCR and speech steps, though participants reported lower temporal demand because the system externalized decision-making. Qualitative themes included restored independence and confidentiality (seven participants had previously compromised ballot secrecy by needing assistance), greater confidence in correct stamping, and appreciation for speech guidance and high-contrast visual output. Participants requested post-stamping confirmation, greater contrast-mode variety, and clearer speech pacing.
Relevance
For accessibility practitioners this paper is notable on several fronts. First, it is a rare empirical demonstration that a carefully-engineered offline, low-cost assistive system can match or exceed expensive ballot-marking devices on usability while avoiding the security and cost critiques that have sunk BMDs in multiple countries. Second, its balanced sample across vision status, braille literacy, and age confronts a persistent problem in accessibility research: tools that work for blind, braille-literate users often fail low-vision, braille-illiterate, or older users, and vice versa. Third, the use of WCAG 2.1 SC 1.4.6 (Contrast Enhanced, 7:1 luminance ratio) as the design target for high-contrast output shows how web accessibility standards can translate directly to hardware-plus-software assistive products. Limitations are clear: the study uses mock Korean ballots in a controlled lab, covers only a single-column format, and does not evaluate long-term use or real polling-station conditions. Practitioners designing accessible voting, form-filling, or any hand-marking workflow should take away the combination of mechanical constraint plus real-time OCR with proximity-based auditory feedback as a generalizable interaction pattern.
Tags: blind and low vision · voting · optical character recognition · multimodal feedback · auditory feedback · visual feedback · assistive technology · civic participation
Standards referenced: WCAG 2.1