The 850nm IR Array.
Standard RGB cameras fail at the edges of computing. They struggle in low light, demand immense host processing, and introduce privacy risks. We engineered a dedicated stereoscopic infrared array to solve optical hand-tracking at the hardware layer.
Structured Light vs RGB Parsing
Unlike Apple Vision Pro which relies on a massive array of RGB cameras and LiDAR, KeyboardHand utilizes a localized 850nm infrared stereoscopic system. By projecting a microscopic dot matrix of structured light across the immediate desk surface, the dual IR sensors can parse depth via parallax distortion instantly.
This means we do not capture visual data. The privacy implications of standard webcams recording user behavior are eliminated. The system mathematically cannot resolve faces or documents—it only sees distance mapping geometry.
Hardware Polling Constraints
- USB Polling Rate 1000 Hz
- Sensor Refresh Cycle 120 FPS
- Inference Delay 1.2 ms
- Total End-to-End Latency 2.4 ms
The Cortex-M7 Brain
Latency is the death of input. If you have to wait for the host CPU to process a frame, interpret a gesture, and translate it to a keystroke, you have already lost the typist's flow state. The ergonomic benefits of optical typing evaporate if the system feels sluggish compared to a mechanical keyboard.
Therefore, KeyboardHand runs all spatial mapping and machine learning inference directly on the device using an embedded ARM Cortex-M7 MCU. To the host operating system (Windows, macOS, Linux), KeyboardHand presents itself as a standard generic USB Human Interface Device (HID). There are no proprietary drivers required for basic functionality, making it ideal for strict enterprise environments.