All-Digital Event-Based Vision Sensor With Multi-Event Generation for Motion/Vibration-Adaptive Detection

Abstract

This article presents an all-digital event-based vision sensor (EVS) using single-photon avalanche diode (SPAD) pixels. The proposed sensor introduces a multi-event generation method that enables the detection of light intensity changes across multiple levels, facilitating advanced event-image processing techniques. The three core processing techniques implemented are event-driven threshold control (EDTC), vibration event noise (VEN) filtering, and adaptive frame-rate control (AFRC). Using the multi-event-based processing, the sensor can dynamically adjust the event-generation threshold (TH), optimize the frame rate, and filter out noise caused by vibrations, thereby enhancing the accuracy of object detection in various challenging environments. EDTC ensures a consistent event rate regardless of the scene’s contrast or lighting conditions, reducing the overall event rate by 55.66%. VEN filtering successfully eliminates 67.25% of the noise generated by vibrations when the sensor module is exposed to constant vibrations, and AFRC optimizes the frame rate based on the relative speed of moving objects, minimizing motion blur and the loss of event data.