A Spike-Based Neuromorphic Architecture of Stereo Vision

Area of Science:

• Computer Vision
• Neuromorphic Engineering
• Biologically-Inspired Computing

Background:

• Stereo correspondence is a challenge for artificial vision, unlike in nature.
• Event-based vision sensors and processors offer potential for efficient stereo matching using temporal information.
• Neuromorphic hardware excels at fast, asynchronous computation leveraging temporal coincidences.

Purpose of the Study:

• To explore the performance benefits of combining event-based sensing with asynchronous, parallel computation for stereo vision.
• To present a hardware spike-based stereo-vision system utilizing neuromorphic computing.
• To quantify the stereo matching performance of the developed system.

Main Methods:

• Interfacing two event-based vision sensors to an event-based mixed-signal neuromorphic processor.
• Designing a prototype interface for stereo-vision emulation on neuromorphic hardware.
• Evaluating stereo matching performance using two distinct datasets.

Main Results:

• Demonstrated a functional hardware spike-based stereo-vision system.
• Quantified the stereo matching performance of the neuromorphic system.
• Showcased the potential of event-based sensing and neuromorphic processing for stereo vision.

Conclusions:

• The developed system provides a viable approach for low-latency stereo vision.
• This work paves the way for end-to-end event-based neuromorphic architectures for stereo vision.
• Highlights the advantages of brain-inspired neuromorphic computing for solving critical vision problems.