Event-Based Vision: A Survey

Area of Science:

• Computer Vision
• Robotics
• Bio-inspired Sensing

Background:

• Event cameras, unlike frame cameras, asynchronously detect per-pixel brightness changes.
• They offer superior temporal resolution (μs), dynamic range (140 dB), low power, and reduced motion blur.
• These properties make them suitable for challenging robotics and computer vision tasks.

Purpose of the Study:

• To provide a comprehensive overview of event-based vision.
• To explore algorithms and applications for event cameras.
• To discuss challenges and opportunities in the field.

Main Methods:

• Review of event camera working principles and available sensors.
• Analysis of algorithms for low-level (feature detection, optic flow) and high-level (reconstruction, recognition) vision tasks.
• Discussion of event processing techniques, including learning-based methods and spiking neural networks.

Main Results:

• Event cameras enable advanced capabilities in robotics and computer vision due to their unique properties.
• A range of algorithms have been developed to process event data for various vision tasks.
• Specialized processors like spiking neural networks are being explored for event-based systems.

Conclusions:

• Event-based vision holds significant potential for enhancing machine perception.
• Further algorithmic development and specialized hardware are needed to fully leverage event camera capabilities.
• The field offers exciting opportunities for more efficient, bio-inspired machine interaction with the world.