Playing classic card games together often fails for blind individuals due to unreadable cards or expensive specialty decks. This mobile web application utilizes offline-capable artificial intelligence for real-time image recognition on the smartphone. It detects playing cards in the camera's field of view instantly and outputs rank and suit with low latency via voice output.
Select one of the playing cards below to hold it in front of the camera. The Edge-ML object detector calculates the bounding boxes. When audio text is enabled, you will hear the spoken output (Web Speech API).
The architecture of image-based inference to ensure low-latency performance on mobile processors.
The neural network was trained on a custom dataset of over 15,000 labeled card photos. The 8-bit quantized version of the model runs directly in the mobile browser using ONNX Runtime Web, saving device resources.
To avoid requiring a cellular connection, the operating system's built-in Web Speech API is used. This enables voice delays of under 50ms – crucial for fluid gameplay at the table.
The web app features a high-contrast dark mode with extremely large touch targets. The navigation is fully compatible with screen readers (TalkBack / VoiceOver) and utilizes vibration haptics for guidance.