Event-based Vision Sensor and On-Chip Processing Development

Abstract

Event-based vision sensors (EVS) detect temporal contrast change with high dynamic range using pixel parallel comparison with a relative threshold and have been proposed as a more efficient means to capture motion information than the existing frame-based RGB sensors. Variety of applications are explored like fast 3D sensing, deblur, video frame interpolation and so on thanks to its high speed and efficient operation. On the other hand EVS generate unexpected event in severe scene environment. The most problematic issue is background detection and flicker detection without any moving object.
This paper shows several event processing to suppress each of undesired background and flicker detection. De-noise and anti-flicker algorithms are characterized through proportion of object preservation ratio and noise suppression ratio evaluated with turning on/off each processing. Filter based processing achieves pretty well denoise characteristics. Local and global flicker is well suppressed by filter-based anti-clicker and edge preservation respectively. Also applicability of each algorithm to target application is suggested.