Dynamic resolution event-based temporal contrast vision sensor

Abstract

This paper presents a dynamic resolution asynchronous event-based dynamic vision sensor(DVS) that benifits from low power consumption. Pixel core in this kind of sensor can be divided into two parts: photoreceptor and motion detection circuit. In this design, power is kept low by turning off some pixels' motion detection circuits when there is no event nearby. The full resolution of this sensor is 64 × 64. There are two working states of the sensor: 1.coarse motion detection and 2.fine detection in the region of interest (ROI). In the first state, each 4×4 pixels are grouped into a pixel block. Their photoreceptors are connect together, sharing only one motion detection circuit to monitor motion in the block region. When a block detects event, the other fifteen pixels in the block are powered up and work individually. Higher resolution image of the ROI is sent out via address event representation (AER) protocol. When no more event is detected in this region, the block goes back to state 1. Simulation result shows that the average power can be reduced to 1/9 of the conventional fixed-resolution DVS. This sensor is fabricated using AMS 0.35μm 4M2P CMOS process. Each pixel contains 3 capacitors and 60 transistors, occupying a silicon area of 34×34μm2, with a fill factor of 15.6%.