A Denoising Method for Event Streams with Graph Attention and Transformer

Event cameras enable high dynamic range and low latency for 3D reconstruction, but indoor noise degrades accuracy. Traditional methods are limited by static thresholds or high computation costs. We propose a GAT-Transformer denoising method that retains event characteristics via spatio-temporal sampling, extracts features with GAT, and classifies events using Transformer. Experiments show lower noise retention and higher signal preservation compared to existing methods, enhancing indoor 3D reconstruction.