摘要
红外焦平面探测器正朝着更大规模、更高帧频、更高集成度的方向发展,这势必带来数据量的急剧增长。动态视觉传感器(Dynamic Vision Sensor,DVS)是一种新型的基于事件触发的视觉感知技术,能够捕捉快速运动物体和瞬间变化的场景细节,具有高动态范围、低延时、数据稀疏、冗余数据少的特点。文中设计了基于红外事件探测器件的数字处理电路及其可视化方法,实现对640×512读出电路事件的数字读出。相较于传统采用仲裁树进行行列仲裁的方式,该设计采用轮询仲裁算法配合仅行仲裁和数据压缩的逻辑实现了对事件的处理,具有仲裁时间短,事件传输效率高的优点,同时具有面向大面阵的可拓展性。采用28 nm制程的FPGA开发板对行为级代码进行验证,同时使用BRAM资源对事件数据按照时间累积,通过HDMI接口高速读出,具象化显示事件成像结果。文中设计等效帧率高于千帧,传输数据吞吐量达0.36 Geps。
Objective With the development of infrared focal plane detector applications,there is an increasing demand for high-frame-frequency imaging,such as the detection and tracking of high-speed small targets,and the tracking and imaging of unmanned aircraft swarms in complex backgrounds,and other fields.The infrared dynamic vision sensor is a bionic sensor different from the traditional infrared sensor,which integrates the photocurrent at a fixed frame rate to form a grayscale image,whereas the infrared dynamic vision sensor asynchronously measures the luminance change of each pixel and outputs the position,time,polarity,etc.of the area of the luminance change in the form of an event stream,which removes the background interference and retains the information of the moving object only,greatly reducing the data Redundancy and lower transmission pressure.Compared with the traditional infrared focal plane readout circuit,the infrared dynamic vision sensor readout circuit has the advantages of high temporal resolution,high dynamic range and low power consumption.Therefore,in this paper,a digital processing circuit based on infrared event detector devices and its visualization method are designed.Methods The paper adopts Round Robin arbitration principle in digital circuit design,which replaces the arbitration tree and reduces the arbitration delay(Fig.4).At the same time,it adopts only row arbitration,together with the output of the whole row information through the compression algorithm,which reduces the delay brought by the column arbitration and the redundancy of the time-stamped information(Fig.5).Two RAM IP cores are used to accumulate the event data over time,enabling writing and reading,and ultimately transmission via the HDMI protocol(Fig.7).Results and Discussions Using the robin arbitration algorithm,the arbitration result can be output in one clock with a delay of 1 clock,which is a significant improvement compared to the arbitration tree structure with a delay of 2-9 clocks.Using only line arbitration with the whole line of information through the compression algorithm output,the equivalent frame rate FR=1085 Hz,data throughput EPS=0.36 G,the results are comparable to the visible light band product parameters(Tab.1).Conclusions Facing the application requirements of infrared focal plane high frame rate readout,a digital processing method for differential infrared dynamic vision sensors is proposed in this paper,which adopts polling arbitration algorithm instead of arbitration tree structure to reduce the impact of arbitration delay;at the same time,it adopts only row arbitration together with the data compression output to reduce the uncertainty and arbitration delay caused by column arbitration.The design is verified using Xilinx FPGA development board,and the corresponding visualization method is proposed.In the visualization process,the 40 bit timestamp information is not used,which can be used for subsequent digital image processing of the event stream.It lays a technical foundation for the development,testing and data processing of the subsequent infrared band dynamic vision sensors.
作者
王政
梁清华
向昊晟
卢天
丁瑞军
WANG Zheng;LIANG Qinghua;XIANG Haocheng;LU Tian;DING Ruijun(National Key Laboratory of Infrared Detection Technologies,Shanghai Institute of Technical Physics,Chinese Academy of Sciences,Shanghai 200083,China;University of Chinese Academy of Sciences,Beijing 100049,China)
出处
《红外与激光工程》
北大核心
2025年第7期122-131,共10页
Infrared and Laser Engineering
基金
中国科学院院重点部署项目。
