Adaptive Neuromorphic Circuit for Stereoscopic Disparity Using Ocular Dominance Map

Area of Science:

• Neuroscience
• Artificial Intelligence
• Computer Engineering

Background:

• Stereopsis, or depth perception, is crucial for visual information processing, computed via binocular disparity.
• Existing real-time disparity computation methods often involve complex, biologically unrealistic algorithms unsuitable for hardware implementation.
• There is a need for simpler, brain-like methodologies to achieve elegant stereopsis in artificial systems.

Purpose of the Study:

• To propose an innovative artificial very-large-scale integration (aVLSI) design for stereopsis.
• To leverage the brain's ocular dominance columnar organization and time-staggered Winner Take All (ts-WTA) for adaptive disparity tuning.
• To create biologically plausible hardware that learns depth information.

Main Methods:

• Developed an aVLSI architecture inspired by the brain's ocular dominance columns.
• Implemented a time-staggered Winner Take All (ts-WTA) mechanism to create disparity-tuned cells.
• Simulated diffusive interactions between disparity cells for stable topological map creation.

Main Results:

• Successfully designed adaptive disparity-tuned cells in silicon, mimicking neural computation.
• Demonstrated the biological plausibility of the approach, aligning with physiological findings on disparity cells.
• Showcased the potential for creating stable, topological disparity maps through cell interaction.

Conclusions:

• The proposed aVLSI design offers a novel and biologically realistic method for hardware-based stereopsis.
• The ts-WTA approach enables adaptive learning of disparities, mirroring neural plasticity.
• This research paves the way for advanced artificial vision systems capable of sophisticated depth perception.