M.N. Adi
Texas State University,
United States
Keywords: AI in built environment, dementia-friendly design, mental health supportive interiors, photo-based risk screening, retrofit decision support
Summary:
Interior environments can unintentionally amplify anxiety, agitation, confusion, and fall risk, especially for people living with dementia or other mental health conditions, yet most practical assessments still depend on time-intensive site visits, specialized expertise, and inconsistent documentation. We present a photo-based AI decision-support workflow that screens everyday interior images (captured by smartphone) for environmental hazards and potential neuropsychological “triggers” and then generates actionable, design-relevant modifications (e.g., lighting adjustments, contrast and wayfinding improvements, furniture layout changes, noise control strategies, and removal/mitigation of common trip hazards). The system combines (1) computer vision to identify objects and spatial cues (e.g., glare-prone luminaires, reflective surfaces, high-clutter zones, ambiguous transitions, low-contrast edges, threshold changes, and obstacle patterns), with (2) a design-logic layer that maps detected features to risk categories and recommended interventions grounded in evidence-informed design principles for safety, aging-in-place, and supportive environments. Outputs are structured for rapid use by designers, facility managers, clinicians, and caregivers: a prioritized checklist, short “why this matters” explanations, and retrofit-oriented recommendations with effort levels (quick fix vs. moderate retrofit vs. renovation-level change). A central contribution is translating AI perception into design-action language rather than generic labels. Instead of only identifying “objects,” the tool explains how specific conditions may increase confusion, stress, sensory overload, or accident likelihood, and proposes targeted modifications that can be implemented in real contexts. The approach is intended to standardize baseline screening across homes, senior living, outpatient behavioral health settings, and campus/workplace environments where mental well-being is a priority. Beyond mainstream practice, this approach can expand access for remote and under-resourced communities and for users facing financial, cultural, or stigma-related barriers to requesting on-site assessments. By enabling low-cost, immediate screening from smartphone photos, the tool can offer a practical “first-pass” evaluation and clear next-step recommendations without requiring travel or specialist availability. The service can be used in an anonymized manner by capturing images with no people present and focusing exclusively on environmental features, allowing individuals, caregivers, and small facilities to obtain guidance while reducing privacy concerns and lowering the threshold for help-seeking. Current privacy practice is operational rather than algorithmic: users are instructed to capture images with no people present, and the workflow is designed to focus on environmental features (lighting, contrast, layout, circulation, clutter, and surfaces). Future versions will explore stronger privacy controls, but the present work emphasizes practical, deployable capture guidance aligned with real-world constraints. Planned evaluation includes expert review (interior design + care specialists) for face validity of risk categories and recommendations; comparison against structured walkthrough assessments to examine agreement, false positives/negatives, and time savings; and usability testing focused on clarity, actionability, and decision confidence. The long-term goal is scalable early screening and continuous improvement of mental-health–supportive environments, enabling preventive interventions before adverse events occur.