AmblyOverlay: An Input-Transparent, Assistive Overlay for Binocular Visual Therapy and Dichoptic Filtering
Pooja Thaker · 2025 · Proceedings of the 27th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS 2025) · doi:10.1145/3663547.3759419
Summary
AmblyOverlay is a proof-of-concept software tool that applies dichoptic color filters to any application window in real time, enabling binocular vision therapy during ordinary computer use — particularly gaming. The system targets people with amblyopia (lazy eye), strabismus (eye misalignment including exophoria), and double vision, conditions that collectively affect around 5% of the world population. Amblyopia alone impacts an estimated 403 million people. Traditional treatments involve patching one eye or using specialized therapeutic games that rely on red-green or red-blue anaglyph glasses to present different visual information to each eye, forcing the brain to coordinate both. However, these therapeutic games are primarily designed for young children and tend to be unengaging for teenagers and adults who prefer modern, fast-paced games. Built with Python, OpenCV, NumPy, and PyQt5, AmblyOverlay captures the content of a selected application window, applies a dichoptic color filter, and renders the result as a transparent, click-through overlay positioned exactly over the original window. This means users can interact normally with their game or application while the therapeutic filter is active. The system supports multiple dichoptic coloring methods: red-cyan anaglyph (fully implemented), grayscale occlusion, and color shift masking (prototyped). A calibration interface with sliders allows adjustment of hue, saturation, and lightness parameters. The developer was personally motivated by their own experience with amblyopia treatment and family members undergoing therapy for exophoria.
Key findings
Performance testing showed minimal overhead from the overlay. In Minecraft, average FPS was 59.8 with the overlay versus 60.0 baseline; in Fortnite (windowed), 58.8 FPS versus 59.6 baseline. The system maintained over 50 FPS consistently across five tested games (Minecraft, Valorant, Fortnite, Geometry Dash, Among Us) on consumer hardware (Intel i7, 16GB RAM, no dedicated GPU required). Resource usage remained under 40% CPU and about 1.5 GB RAM. AI-based adaptive filtering using object detection models (YOLOv8, Mask2Former) was explored but proved too slow for real-time use — inference times ranged from 166ms to 1834ms per frame depending on model and resolution, making it impractical for interactive gameplay. This feature was deferred to future work pending GPU acceleration and model optimization. Preliminary consultations with three eye care professionals (ophthalmologists and optometrists) validated the concept and highlighted the importance of display-specific calibration, conservative usage limits (around 30 minutes per day), and adjustable filter intensity. No formal clinical trials or user studies have been conducted yet.
Relevance
AmblyOverlay addresses a genuine gap in accessibility: existing dichoptic therapy tools are tied to specific games or proprietary hardware, making them inflexible and unappealing to older patients. By working as a system-wide overlay compatible with any windowed application, it lets patients choose their own content — a significant motivational advantage for long-term therapy adherence. The transparent, click-through design is the key technical innovation, allowing therapy to integrate seamlessly into normal computer use rather than requiring dedicated sessions. However, this is currently a technical proof-of-concept with no clinical validation. The system is Windows-only, only red-cyan filtering is fully implemented, and calibration is not yet connected to the live filter output. Practitioners should note the potential but await clinical trial results before considering it a validated therapeutic tool. The work highlights an underserved population — teens and adults with binocular vision disorders who find existing therapy tools too childish or restrictive.
Tags: amblyopia · visual therapy · dichoptic vision · assistive technology · binocular vision · game accessibility · computer vision