Event cameras are neuromorphic sensors that capture asynchronous and sparse event stream when per-pixel brightness changes. The state-of-the-art processing methods for event signals typically aggregate events into a frame or a grid. However, events are dense in time, these works are limited to local information of events due to the stacking. In this paper, we present a novel spatiotemporal representation learning method which can capture the global correlations of all events in the event stream simultaneously by tensor decomposition. In addition, with the events are sparse in space, we propose an Elastic Net-incorporated tensor network (ENTN) model to obtain more spatial and temporal details about event stream. Empirically, the results indicate that our method can represent the spatiotemporal correlation of events with high quality, and can achieve effective results in applications like filtering noise compared with the state-of-the-art methods.

翻译：事件相机是一种神经形态传感器，可在每像素亮度变化时捕获异步稀疏的脉冲流。现有主流的脉冲信号处理方法通常将脉冲聚合为帧或网格结构。然而，由于脉冲在时间上密集分布，这些方法受限于堆叠操作导致的局部信息提取。本文提出一种新型时空表示学习方法，通过张量分解能够同时捕获脉冲流中所有事件的全局相关性。此外，针对脉冲在空间上的稀疏特性，我们提出一种弹性网络融入张量网络模型，以获取脉冲流更丰富的时空细节。实验结果表明，该方法能高质量地表征脉冲的时空关联性，并在噪声滤波等应用中取得优于现有方法的效果。