Event-Based Vision
Overview
Event-based vision studies cameras that report changes in brightness asynchronously rather than capturing fixed-rate images, making them useful for fast motion, low light, and high-dynamic-range scenes. Methods convert events into time surfaces, voxel grids, or continuous-time features, then apply neural networks, spiking models, or geometric estimation for tasks such as detection, optical flow, depth, and visual odometry. Work also combines events with RGB, inertial, or LiDAR data and evaluates systems on benchmarks such as DSEC, MVSEC, Gen1, and DVS Gesture. A central challenge is making results robust to sensor noise and hardware differences while balancing accuracy, latency, energy use, and performance against frame-based methods in real-world settings.
Latest papers
232 papers20 papers in the last four weeks, down 26% on the four weeks before. 0.1% of all new papers.