A Compact VLSI System for Bio-Inspired Visual Motion Estimation.

Area of Science:

• Computational Neuroscience
• Computer Vision
• VLSI Design

Background:

• Existing visual motion estimation algorithms often require significant computational resources.
• Mimicking biological visual systems offers a promising approach for efficient motion perception.
• Low-cost embedded systems demand optimized algorithms and hardware architectures.

Purpose of the Study:

• To propose a bio-inspired visual motion estimation algorithm based on motion energy.
• To develop a compact Very-Large-Scale Integration (VLSI) architecture for real-time implementation on low-cost embedded systems.
• To introduce a confidence map for assessing the reliability of motion estimation.

Main Methods:

• Algorithm development inspired by primate retinal, V1, and MT neuron functions.
• Implementation of ternary edge extraction, spatiotemporal filtering, motion energy extraction, and velocity integration.
• Design of a VLSI architecture featuring multi-level pipelining and parallel processing arrays optimized for hardware efficiency.

Main Results:

• A prototype achieved real-time performance (30 frames/s) for velocity estimation on a Field-Programmable Gate Array (FPGA) platform (Zynq 7020).
• The system demonstrated high accuracy with average endpoint error < 0.5 pixels and angular error < 10°.
• The algorithm's reliance on additions and multiplications facilitates low-cost hardware implementation.

Conclusions:

• The proposed bio-inspired algorithm and VLSI architecture offer an efficient solution for real-time visual motion estimation.
• The system is suitable for low-cost embedded applications requiring accurate motion perception.
• The confidence map provides valuable information on the reliability of the estimation results.