Emerging Trends and Applications of Neuromorphic Dynamic Vision Sensors: A Survey
Hadi Aliakbarpour, Ahmad Moori, Javad Khorramdel, Erik Blasch, Omar Tahri
Abstract
Traditional frame-based cameras, foundational in computer vision, face challenges, such as motion blur, restricted dynamic range, and limited temporal resolution, which limit their use in various scenarios. Neuromorphic dynamic vision sensors (DVSs), inspired by biological vision systems, offer a significant shift with exceptional temporal resolution and pixel-level detection of light intensity changes, generating a continuous stream of events that encode time, location, and polarity. DVSs surpass traditional cameras with a dynamic range of 140 dB (compared to 60 dB), resilience to uneven lighting, minimal latency, ultra-high temporal resolution equivalent to over 100 000 frames per second, and low power consumption of just 1 mW. These capabilities mark DVSs as a transformative advancement in sensing technology. However, DVSs demand novel processing techniques due to their unique data output. This survey revisits both overlooked and newly available research, categorizing established and emerging techniques. It explores a range of applications, from autonomous systems, vehicle navigation, object detection, and depth estimation to gesture recognition. In addition, the survey introduces newer applications, such as super-resolution, rolling shutter correction, robotic manipulation, data compression, and space situational awareness. This article also tackles key challenges, such as DVS calibration, frame reconstruction, and simulation. To support future advancements, it provides a comprehensive overview of simulation platforms and the latest datasets, aiming to inspire further research and promote wider adoption of DVS technology in rapidly evolving fields.