Neuromorphic Stereo Vision: A Survey of Bio-Inspired Sensors and Algorithms

Area of Science:

• Robotics
• Computer Vision
• Computational Neuroscience

Background:

• Visual sensors map 3D environments to 2D representations, necessitating depth recovery.
• Stereo vision, using multiple perspectives, is a common method for depth perception in both biological and artificial systems.
• Current artificial stereo vision algorithms face challenges with high energy consumption and latency, particularly in solving the correspondence problem.

Purpose of the Study:

• To investigate event-based sensors and algorithms for stereo vision.
• To develop more biologically plausible and efficient depth perception systems for robots.
• To leverage the temporal constraint offered by event-based cameras and Spiking Neural Networks.

Main Methods:

• Utilizing event-based cameras that capture visual information asynchronously based on pixel-level changes.
• Employing Spiking Neural Networks (SNNs) that process information based on the timing of neural spikes.
• Focusing on binocular stereo vision to recover depth information from two synchronized camera inputs.

Main Results:

• Event-based stereo vision, combined with SNNs, offers a promising approach to overcome limitations of traditional stereo vision methods.
• The temporal dynamics captured by event-based sensors can be effectively exploited by SNNs to address the correspondence problem.
• This approach has the potential to significantly reduce energy demands and latency in artificial vision systems.

Conclusions:

• Event-based stereo vision represents a significant advancement towards creating robots with more natural and efficient visual processing capabilities.
• The integration of event-based cameras and SNNs offers a pathway to more biologically inspired artificial intelligence.
• Further research in this area can lead to breakthroughs in autonomous systems and robotics.