Inspiration
Total blindness affects millions globally, isolating individuals from visual environments and color perception. Our inspiration stemmed from a desire to bypass damaged ocular pathways using non-invasive neurostimulation, translating digital camera frames directly into bio-compatible electrical frequency signatures that the brain can interpret safely.
What it does
The Bio-Visual Neural Core is a wearable edge-AI assistive system that captures real-time optical streams via a compact camera. It executes local image processing and color breakdown (RGB values) using TensorFlow Lite, converting pixel matrices into precise, deterministic micro-current electrical frequencies (PWM signals). These signals are delivered safely via non-invasive cranial electrodes positioned over the mastoid process behind the ears to restore spatial and color perception.
How we built it
- Computer Vision: Python and OpenCV for real-time video capture, spatial pixel parsing, and color extraction.
- Edge Intelligence: TensorFlow Lite (TinyML) running locally on device for efficient neural decoding without cloud latency.
- Firmware & Control: C++ (Arduino IDE) running on an ATmega328P microcontroller (Arduino Nano) with rigorous packet boundary verification and voltage/current regulation circuits.
- Safety Architecture: Integrated hardware boundaries ensuring micro-ampere safety limits and isolated power rails.
Challenges we ran into
- Maintaining real-time frame processing speeds while running local neural inference on resource-constrained hardware.
- Guaranteeing absolute electrical safety and stability for non-invasive stimulation over the mastoid process.
- Developing a deterministic mapping algorithm to reliably translate complex RGB color gradients into distinct bio-compatible frequency telemetry blocks.
Accomplishments that we're proud of
- Successfully engineering a complete, end-to-end local Edge AI pipeline that operates entirely offline with complete user privacy.
- Implementing a robust safety-first electrical regulation framework for non-invasive neural stimulation.
- Creating a cost-effective, highly accessible hardware prototype built on an ultra-low-cost single-MCU architecture.
What we learned
- The immense potential of combining TinyML with non-invasive bio-electric interfaces to solve critical accessibility challenges.
- The importance of strict local data processing to safeguard user privacy in medical and assistive technologies.
What's next for Bio-Visual Neural Core
- Optimizing neural mapping algorithms for enhanced color resolution and depth perception.
- Conducting rigorous algorithmic validation and expanding clinical collaboration frameworks.
- Scaling the hardware architecture into a comfortable, low-profile wearable form factor for everyday independent use.
Log in or sign up for Devpost to join the conversation.