低维流形约束下的事件相机去噪算法

A Denoising Algorithm for Event Cameras Based on Low-Dimensional Manifold Constraint

本文主要提出一个新的基于流形约束的事件相机去噪算法。事件相机是一类新型的视觉传感器,以高时间分辨率(微秒)感知场景亮度变化,同时输出具有像素位置、时间及极性的事件流。事件相机在传输亮度变化的同时受到噪声的干扰,带噪的事件流会对后续的应用造成不利的影响。为了解决该问题,本文假设事件分布在高维空间中的低维流形上,利用事件点间相似信息建立图模型以近似流形结构,结合图的流形平滑约束,实现事件流去噪。该算法首次将基于图的流形约束引入事件去噪问题中并且直接处理连续的事件序列。仿真实验和真实数据结果证明了事件去噪算法的有效性。

This paper proposes a new denoising algorithm for event cameras based on manifold constraints.An event camera is a new type of vision sensor,which can perceive the change of scene brightness with high temporal resolution(microseconds)and output the event stream with pixel position,time and polarity.Event cameras are disturbed by noise while transmitting brightness changes,and noisy event streams will adversely affect subsequent applications.In order to solve this problem,this paper assumes that events are distributed on a low-dimensional manifold in high-dimensional space,uses the similarity information between event points to build a graph to approximate the manifold structure,and combines the manifold smoothing constraint of graph to complete the event stream denoising.For the first time,the proposed algorithm introduces the graph-based manifold constraint into the event denoising problem and directly processes successive event sequences.Experiments on simulated data and real datasets demonstrate the effectiveness of the event denoising algorithm.