Case Study: In-the-Field Accessibility Information Collection Using Gamification
Akihiro Miyata, Kazuki Okugawa, Yusaku Murayama, Akihiro Furuta, Keihiro Ochiai, Yuko Murayama · 2023 · Proceedings of the 20th International Web for All Conference (W4A '23) · doi:10.1145/3587281.3587288
Summary
This study introduces and evaluates a crowdsourcing platform designed to collect real-world accessibility information for constructing accessibility maps that support people with mobility disabilities. Accessibility maps are critical for safe navigation by wheelchair users and others with mobility impairments, but creating them through professional audits is expensive and limits coverage. While automatic auditing (using computer vision on streetscape imagery) and crowdsourced virtual auditing (using Google Street View) have been proposed as alternatives, both suffer from outdated photos, occlusions, and gaps in traffic-free zones where image-capturing vehicles cannot enter. The researchers designed a smartphone platform with four distinct modes to accommodate users with varying levels of free time and motivation. "Reporter" mode targets high-time, high-motivation users who manually photograph and label accessibility barriers. "Gaming Reporter" adds gamification through a monster-collecting mechanic — users photograph accessibility problems to collect monsters whose appearance reflects the barrier type (e.g., a monster with stairs on its body for stair barriers), with a ResNet-50 image classifier (F-measure 0.95) automatically categorizing the barrier. "Walker" mode targets high-motivation but time-poor users, passively collecting inertial sensor data from smartphones carried in pockets during regular walking. "Gaming Walker" gamifies the passive sensing approach through a team-based territory competition game where walking expands virtual territory. A 1D convolutional neural network (F-measure 0.99) classifies road surface types from accelerometer and gyroscope data into seven categories including flat, up-step, down-step, up-stairs, down-stairs, up-high slope, and down-high slope.
Key findings
An eight-week experiment with 28 university students (ages 18-24) using a Latin square design across four two-week periods confirmed two key hypotheses. First, without gamification, participation differed significantly between high- and low-motivation participants — highly motivated people contributed more. Second, with gamification, this motivation gap was substantially reduced; low-motivation participants increased their contributions to levels comparable with high-motivation users. Qualitative analysis revealed that "fun" was the primary reason for using gamified modes, while "ease of operation" rather than "contribution to others" drove use of non-gamified modes. The most common reason for non-use was busyness, except for the Walker mode where "little feedback" was the top complaint, highlighting a design tension between minimal-effort modes and users' need for a sense of accomplishment. Critically, neither the posting mode nor motivation level had a significant effect on data quality — gamification improved participation quantity without compromising the quality of collected accessibility information. Posting patterns also differed by mode: passive sensing modes (Walker/Gaming Walker) were used during morning commutes, while active reporting modes (Reporter/Gaming Reporter) were used in late afternoon and evening.
Relevance
This research addresses a fundamental challenge in accessibility infrastructure: the gap between the need for comprehensive, up-to-date accessibility information about the built environment and the cost of collecting it. The four-mode platform design offers a practical template for organizations seeking to build accessibility maps through citizen participation — acknowledging that contributors have different time constraints and motivation levels rather than assuming a one-size-fits-all approach. The finding that gamification closes the motivation gap without degrading data quality is particularly actionable for developers of accessibility crowdsourcing tools. However, all participants were non-disabled university students aged 18-24, which limits generalizability — wheelchair users and older adults may have different motivations and preferences. The study also highlights an important design tension: passive, low-effort collection modes need meaningful feedback mechanisms to sustain engagement, while game-based modes may not appeal to all demographics.
Tags: crowdsourcing · gamification · accessible maps · physical accessibility · pedestrian infrastructure · deep learning · urban accessibility · mobile application · wheelchair accessibility · computer vision