cd ~/

Speckle-Based Eye Tracking: A Novel Approach

Using laser speckle patterns for high-precision eye tracking - the physics, implementation, and path to a patent.

Evyatar Bluzer
2 min read

We've been exploring an unconventional approach to eye tracking that exploits laser speckle patterns. It's now mature enough to file a patent.

What is Speckle?

When coherent light (laser) illuminates a rough surface, the scattered light creates an interference pattern of bright and dark spots called "speckle." This pattern is:

  • Unique to the surface: Like a fingerprint
  • Sensitive to viewing angle: Changes with observer position
  • High spatial frequency: Fine-grained structure

Speckle on the Eye

The human cornea and iris have surface texture that creates speckle when illuminated with coherent NIR light.

Key insight: as the eye rotates, the speckle pattern translates.

If we can track the speckle pattern's movement, we can track eye rotation with very high precision.

Implementation Concept

VCSEL Illumination → Corneal Reflection → Speckle Pattern →
High-Speed Camera → Pattern Correlation → Gaze Estimation

Illumination

Vertical-Cavity Surface-Emitting Lasers (VCSELs):

  • Coherent enough for speckle generation
  • Eye-safe at low power
  • Compact and efficient

Imaging

Capture speckle pattern on specialized camera:

  • Small pixel size to resolve speckle (≤3μm)
  • High frame rate for fast eye movements (>200Hz)
  • NIR sensitivity matching VCSEL wavelength

Pattern Tracking

Cross-correlation between consecutive frames:

offset = argmax(correlation(frame_n, frame_n-1))
rotation = f(offset, geometry)

Sub-pixel correlation gives sub-degree rotation precision.

Advantages Over Traditional Eye Tracking

Resolution: Traditional glint/pupil tracking limited by image resolution. Speckle provides sub-pixel precision.

Speed: Pattern correlation is computationally simpler than feature detection.

Robustness: Less sensitive to pupil dilation, partial occlusion.

Challenges

Speckle decorrelation: If surface changes (tears, blinking), pattern changes. Need to detect and handle decorrelation events.

Hardware requirements: Needs coherent source and high-resolution sensor - adds cost and power.

Eye safety: Coherent radiation requires careful safety analysis. Must meet IEC 62471 Class 1.

Patent Strategy

We're filing on:

  1. Method of eye tracking using speckle correlation
  2. Apparatus with specific VCSEL + sensor configuration
  3. Decorrelation detection and recovery

The combination of eye safety constraints, speckle imaging requirements, and correlation algorithms is novel.

Path Forward

Current status: bench prototype showing 0.1° accuracy, limited to slow movements.

Needed for product:

  • Faster correlation (GPU/DSP implementation)
  • Reliable decorrelation handling
  • Integration with display optics

This may not make V1 product, but it's promising for future iterations.

[Patent granted 2020: US10948981 "Method and System for Eye Tracking Using Speckle Patterns"]

Comments