Event Stream Denoising Method Based on Spatio-Temporal Density and Time Sequence Analysis.

Area of Science:

• Neuromorphic Engineering
• Computer Vision
• Signal Processing

Background:

• Event cameras offer high dynamic range, temporal resolution, and low power consumption, mimicking human retinal imaging.
• Event streams often contain significant noise due to hardware/software factors, rendering traditional denoising methods ineffective.
• Robust noise reduction is crucial for leveraging event camera capabilities.

Purpose of the Study:

• To develop an event stream noise reduction and visualization algorithm robust to various noise types.
• To enhance the clarity and coherence of event frames generated from noisy event streams.
• To address the limitations of existing denoising techniques for event camera data.

Main Methods:

• A two-stage filtering approach: initial spatio-temporal density-based filtering for background/additive (BA) noise, followed by fine filtering.
• Fine filtering employs temporal sequence analysis of event pixels and their neighbors to remove hot noise.
• A visualization algorithm adaptively overlaps events based on density differences for frame coherence.

Main Results:

• The proposed algorithm effectively reduces noise in event streams from real-world scenes and public datasets.
• Clear and coherent event frames are successfully generated, even under varying event densities and noise levels.
• Experimental validation confirms the algorithm's robustness and effectiveness in denoising and visualization.

Conclusions:

• The developed algorithm provides an effective solution for noise reduction in event camera data.
• The method enhances the quality of event streams, enabling more reliable downstream applications.
• This work contributes to the practical usability of event cameras in diverse conditions.