-
仿人眼视网膜成像包括两方面:(1)结构方面,实现近似视网膜感光细胞的变分辨布局,即中央凹高分辨、周边低分辨;(2)机理方面,在(1)的基础上,形成近似笛卡尔到对数极坐标的映射。根据文献,将仿人眼视网膜成像实现方法划分为软件、电路、光学三类。其中,基于软件可以通过常规图像传感器得到一幅图像,然后采用适当的算法转换成对数极坐标图像[7-8],此种方式具有操作简便、成本低的优点,缺点在于实时性不明显,完全依赖于处理器能力。一些通过FPGA[9-10]实现类似上述功能的方法本质上也属于基于软件方式。光学方式主要是通过对光学镜头中(固定[11]或移动[12-13])的光学组件进行优化,使得某一特定区域实现高分辨率成像,从而达到变分辨成像目的。电路方式可以根据主被动成像进一步细化,较为典型的被动式仿人眼视网膜成像是通过CCD或CMOS图像传感器像素非均匀化实现,例如:1990年,意大利热那亚大学Giulio Sandini等人[14],基于CCD加工技术,研制了单元离散分布的首个仿人眼视网膜传感器。像素所处的半径随视场离心率线性增加,像素大小随着半径增加,如图1(a)所示。传感器外围包括30圆环,每环64个,总共1 920个像素点,像素大小从30 μm逐渐增大到412 μm,中央凹区成菱形形状,用于弥补中央凹区域太小而无法映射的缺陷,包括102个像素,尺寸11 mm×11 mm。与CCD工艺相比较,CMOS在数据存储上可控性更高,更适合于模拟人眼视网膜信息获取,更加容易与微处理器实现对接,典型结构如图1(b)所示,由校际微电子中心(IMEC)研发的仿人眼图像传感器[15],基于0.7 μm的CMOS工艺,像元尺寸14 μm,包含了8 013个像素,其中的7 168(56环,每环128)个像素属于对数极坐标转换像素,中央凹处覆盖的像素为845个,传感器尺寸8.1 mm。之后基于0.35 μm CMOS加工工艺,实现了传感器外围110环,每环252个,共27 720个像素,中央凹区共5 773个像素,整个传感器总共33 493个像素,最外环像素大小与最内环像素大小比例为17,整个传感器尺寸与最小像素尺寸比值为1 100,整个传感器尺寸为7.1 mm[16]。北京理工大学曹峰梅等人[17-18]完成了国内首个仿人眼视网膜图像传感器,如图1(c)所示,感光面尺寸10.226 mm,像元尺寸14 μm×14 μm,证明了其可以有效消除图像旋转产生的影响。由于集成电路设计复杂、非均匀探测器存在串扰等问题,也可以通过透镜阵列实现仿人眼视网膜布局,而各探测器仍需要采用相同尺寸,以此提高光能利用率[19]。
主动成像方式因环境适应性更强,近年来也得到深入发展,较为典型的激光三维成像雷达,已广泛应用于无人驾驶、避障、地形勘测等领域。主动成像系统包括发射与接收两部分,因此,结合仿人眼视网膜机理可从发射与接收两个方面形成仿人眼视网膜成像。值得注意的是,目前,激光三维成像雷达多以雪崩二极管(APD)或小面阵APD阵列作为接收核心器件,相比较CCD或CMOS工艺,APD因受噪声串扰影响仍难以形成大面阵芯片(国外可实现最大面阵256×256[20],国内为32×32[21])。因此,主动成像仿人眼视网膜在前端更容易实现,即通过变分辨扫描实现。例如:加州大学洛杉矶分校的Jiang等人[22],提出采用时间延展的方法实现时间飞行扫描的激光雷达,通过在不同光谱上实现针对固定区域的高分辨率采集,扫描频率达到1 MHz,达到仿人眼视网膜成像效果。北京航空航天大学陈伟海等人[23],提出利用DSP+FPGA的硬件平台构架,对环境进行分步扫描,在线规划下一步的扫描规律,以减少无用信息获取。南京理工大学顾国华等人[24]利用背景与目标的非连续性特点,提出了自适应的单光子计数三维成像激光雷达,基于扫描成像方式,使得数据采样效率提高了85%。北京理工大学曹杰、郝群等人[25]以MEMS光学器件为核心,结合单像素探测器,完成了类似视网膜环形变分辨扫描与采样,实现了仿人眼视网膜机理成像,满足旋转与尺度不变特性。近年来,对于以单像素探测器为基础的量子关联成像方法(也被称为鬼成像),随着计算机能力大幅提升而得到进一步研究,其中,英国格拉斯哥大学Phillips等人[26]提出利用人眼视网膜变分辨成像特点,实现变分辨率成像,在相同分辨率下,通过DMD生成疏密程度非均匀的二维随机散斑光场,视场中央模拟人眼中央凹的高分辨率成像,周边区域模拟人眼外围低分辨率成像,以适应不同的场景。将上述方式进行总结,如表1所示,对于各种方式的具体应用将在实例化应用中进一步介绍。
表 1 实现方式比较
Table 1. Comparison of implementation
Implementation items Circuit Optics Software Passive Active Core method Non-uniform pixels Non-uniform scanning speckles Variable resolution design of optical components Logarithmic polar correlation algorithms Key feature Compression when sampling, high resolution in gaze imaging region +peripheral low resolution imaging sampling Achieve variable resolution with image sensors Compressing after sampling Advantages Good real-time performance, high sensitivity and low data volume Achieve local high resolution with good scalability Structures are simple and easy to connect with existing equipment Disadvantage Nonuniform response No reduction in data volume Poor real-time performance Difficulty Difficult Medium Medium Easy
Research progress of bio-inspired retina-like imaging (Invited )
-
摘要: 人眼视网膜因具有变分辨成像及冗余数据压缩的优势为成像方法与应用开辟了新途径。基于人眼视网膜成像机理,有效解决了大视场、高分辨、实时性难以兼顾的问题。随着半导体制造工艺与计算机处理能力的迅猛发展,仿人眼视网膜成像系统已广泛应用于视觉导航、识别与跟踪、生物医疗等领域, 目前,正朝着小型化、高效化、智能化方向发展。对比分析了仿人眼视网膜成像实现方法的优缺点,以现有实例化应用分析了仿人眼视网膜成像技术的发展现状并提出挑战与机遇,为进一步研究仿人眼视网膜成像提供参考与借鉴。Abstract: With prominent advantages of rotation and scaling invariance and decreasing redundant information, etc., the features of human retina provides novel approaches for optical imaging and applications. The issue on the relationship among large filed of view, high resolution, real-time ability is well solved by the use of retina-like imaging mechanism. The retina-like imaging systems are widely used in the fields such as visual navigation, recognition and tracking, biomedical engineering. Meanwhile, the systems become small, efficiency and intelligent due to the fast development of semiconductor processing and computer processing techniques. The implementations of retina-like systems were compared, and the merit and demerit were summarized in the paper. Some typical practical applications were taken as example for presenting current research status. Opportunities and challenges were discussed for studying bio-inspired retina-like imaging further.
-
表 1 实现方式比较
Table 1. Comparison of implementation
Implementation items Circuit Optics Software Passive Active Core method Non-uniform pixels Non-uniform scanning speckles Variable resolution design of optical components Logarithmic polar correlation algorithms Key feature Compression when sampling, high resolution in gaze imaging region +peripheral low resolution imaging sampling Achieve variable resolution with image sensors Compressing after sampling Advantages Good real-time performance, high sensitivity and low data volume Achieve local high resolution with good scalability Structures are simple and easy to connect with existing equipment Disadvantage Nonuniform response No reduction in data volume Poor real-time performance Difficulty Difficult Medium Medium Easy -
[1] 李晗, 段海滨, 李淑宇. 猛禽视觉研究新进展[J]. 科技导报, 2018, 36(17): 52-67. Li Han, Duan Haibin, Li Shuyu. New progress in raptor vision research [J]. Science & Technology Review, 2018, 36(17): 52-67. (in Chinese) [2] Floreano D, Pericet-Camara R, Viollet S, et al. Miniature curved artificial compound eyes [J]. Proc Natl Acad Sci U S A, 2013, 110(23): 9267-9272. doi: 10.1073/pnas.1219068110 [3] 胡雪蕾, 高明, 陈阳. 大视场曲面仿生复眼光学系统设计[J]. 红外与激光工程, 2019, 49(1): 0114002. doi: 0114002 Hu Xuelei, Gao Ming, Chen Yang. Design of curved bionic compound eye optical system with large field of view [J]. Infrared and Laser Engineering, 2019, 49(1): 0114002. (in Chinese) doi: 0114002 [4] Cheng Yang, Cao Jie, Zhang Yangkun, et al. Review of state-of-the-art artificial compound eye imaging systems [J]. Bioinspiration & Biomimetics, 2019, 14: 0310023. [5] Schwartz E L. Spatial mapping in primate sensory projection - analytic structure and relevance to perception [J]. Biological Cybernetics, 1977, 25(4): 181-194. doi: 10.1007/BF01885636 [6] Javier Traver V, Alexandre Bernardino. A review of log-polar imaging for visual perception in robotics [J]. Robotics and Autonomous Systems, 2010, 58(4): 378-398. doi: 10.1016/j.robot.2009.10.002 [7] Paolo P, Gamba Ele, Lombardi Luca, et al. Log-map analysis [J]. Parallel Computing, 2008, 34(12): 757-764. doi: 10.1016/j.parco.2008.09.008 [8] Liu Zhaoqin, Peng Man, Di Kaichang, et al. A continuative variable resolution digital elevation model for ground-based photogrammetry [J]. Computers & Geosciences, 2014, 62: 71-79. [9] 韩溟, 郭峰, 王道平. 图像对数极坐标变换的FPGA实现[J]. 弹箭与制导学报, 2011, 31(6): 44-46. doi: 10.3969/j.issn.1673-9728.2011.06.013 Han Ming, Guo Feng, Wang Daoping. The implementation of LPT for image by FPGA [J]. Journal of Projectiles, Rockets, Missiles and Guidance, 2011, 31(6): 44-46. (in Chinese) doi: 10.3969/j.issn.1673-9728.2011.06.013 [10] Wong Wai Kit, Choo Chee Wee, Loo Chu Kiong, et al. FPGA implementation of log-polar mapping[C]//IEEE 15th International Conference on Mechatronics and Machine Vision in Practice, 2008: 45-50. [11] Wick D, Martinez T, S Restaino, et al. Foveated imaging demonstration [J]. Optics Express, 2002, 10(1): 60-65. doi: 10.1364/OE.10.000060 [12] Hao Qun, Wang Zihan, Cao Jie, et al. A hybrid bionic image sensor achieving fov extension and foveated imaging [J]. Sensors, 2018, 18: 10424. [13] Niu Yajun, Chang Jun, Lv Fengxiang, et al. Low-cost dynamic real-time foveated imager [J]. Applied Optics, 2017, 56(28): 7915. doi: 10.1364/AO.56.007915 [14] Massimo Tistarellia, Giulio Sandinia. Estimation of depth from motion using an anthropomorphic visual sensor [J]. Image and Vision Computing, 1990, 8(4): 271-278. doi: 10.1016/0262-8856(90)80003-C [15] Sandini G, Questa P, Scheffer D, et al. A retina-like CMOS sensor and its applications[C]//Proceedings of the 2000 IEEE Sensor Array and Multichannel Signal Processing Workshop, 2000: 514-519. [16] Pardo F, Boluda J A, Pérez J J, et al. Response properties of a foveated space-variant CMOS image sensor[C]//1996 IEEE International Symposium on Circuits And Systems, ISCAS'96. 1996: 373-376. [17] Cao Fengmei, Lin Yabin, Bai Tingzhu, et al. Image rotation-elimination based on a retina-like sensor [J]. Optical Engineering, 2015, 54: 12310512. [18] 曹峰梅, 牟勇, 白廷柱. 基于仿视网膜成像器件的折反射全景系统设计[J]. 光学技术, 2018, 44(2): 133-139. Cao Fengmei, Mou Yong, Bai Tingzhu. Design of a catadioptric panorama system based on retina-like imaging device [J]. Optical Technique, 2018, 44(2): 133-139. (in Chinese) [19] Hao Qun, Meng Lingtong, Cheng Yang, et al. Design of a retina-like transmitting system based on a curved lens array [J]. Applied Optics, 2018, 57(29): 8494-8502. doi: 10.1364/AO.57.008494 [20] Brian F Aull, Duerr Erik K, Frechette Jonathan P, et al. Large-format Geiger-mode avalanche photodiode arrays and readout circuits [J]. IEEE Journal of Selected Topics in Quantum Electronics, 2018, 24: 38005102. [21] 孙剑峰, 姜鹏, 张秀川, 等. 32×32面阵InGaAs Gm-APD激光主动成像实验[J]. 红外与激光工程, 2016, 45(12): 1206006. doi: 10.3788/IRLA201645.1206006 Sun Jianfeng, Jiang Peng, Zhang Xiuchuan, et al. Experimental research of 32×32 InGaAs Gm-APD arrays laser active imaging [J]. Infrared and Laser Engineering, 2016, 45(12): 1206006. (in Chinese) doi: 10.3788/IRLA201645.1206006 [22] Jiang Yunshan, Karpf Sebastian, Jalali Bahram. Time-stretch LiDAR as a spectrally scanned time-of-flight ranging camera [J]. Nature Photonics, 2020, 14(1): 14. doi: 10.1038/s41566-019-0548-6 [23] 陈伟海, 宋蔚阳, 荣利霞, 等. 仿人眼功能的三维激光扫描算法[J]. 北京航空航天大学学报, 2009, 3(5): 563-566. Chen Weihai, Song Weiyang, Rong Lixia, et al. 3D laser scanning algorithm with human-like eye function [J]. Journal of Beijing University of Aeronautics and Astronautics, 2009, 3(5): 563-566. (in Chinese) [24] Ye Ling, Gu Guohua, He Weiji, et al. Adaptive target profile acquiring method for photon counting 3-D imaging lidar [J]. IEEE Photonics Journal, 2016, 8(6): 2625801. doi: 10.1109/JPHOT.2016.2625801 [25] Cao Jie, Hao Qun, Xia Wenze, et al. Design and realization of retina-like three-dimensional imaging based on a MOEMS mirror [J]. Optics and Lasers in Engineering, 2016, 82: 1-13. [26] David B Phillips, Sun Mingjie, Taylor Jonathan M, et al. Adaptive foveated single-pixel imaging with dynamic supersampling [J]. Science Advances, 2017, 3(4): 1-10. [27] Cheon W Shin, Inokuchi Seiji, Kim Kwang I. Retina-like visual sensor for fast tracking and navigation robots [J]. Machine Vision and Applications, 1997, 10(1): 1-8. doi: 10.1007/s001380050053 [28] 乔良, 鲍泓, 郑颖. 无人驾驶汽车决策控制技术研究: 基于驾驶态势图[C]//中国计算机用户协会网络应用分会2017年第二十一届网络新技术与应用年会论文集, 2017: 293-296. Qiao Liang, Bao Hong, Zheng Ying. Research on autonomous driving decision and control technology: Based on driving state map[C]//Proceedings of the 21st Annual Meeting of New Network Technology and Application in 2017 of Network Application Branch of China Computer Users Association, 2017: 293-296. (in Chinese) [29] 范晋祥, 张渊, 王社阳. 红外成像制导导弹自动目标识别应用现状的分析[J]. 红外与激光工程, 2007, 36(6): 778-781. doi: 10.3969/j.issn.1007-2276.2007.06.002 Fan Jinxiang, Zhang Yuan, Wang Sheyang. Analysis of application status of ATR technologies in infrared imaging guided missiles [J]. Infrared and Laser Engineering, 2007, 36(6): 778-781. (in Chinese) doi: 10.3969/j.issn.1007-2276.2007.06.002 [30] 赵刚, 李嵘, 王敏. 符合人眼视觉特性的前背景匹配快速算法[J]. 计算机工程与应用, 2008, 44(20): 80-82. doi: 10.3778/j.issn.1002-8331.2008.20.025 Zhao Gang, Li Rong, Wang Min. Fast foreground/background matching algorithm based on human visual properties [J]. Computer Engineering and Applications, 2008, 44(20): 80-82. (in Chinese) doi: 10.3778/j.issn.1002-8331.2008.20.025 [31] 李言俊. 视觉仿生成像制导技术及应用[M]. 北京: 国防工业出版社, 2006. Li Yanjun. Visual Imitation Image Guidance Technology and Application[M]. Beijing: National Defense Industry Press, 2006. (in Chinsese) [32] 粟宇路, 曹峰梅, 王凡, 等. 仿视网膜探测器在相向运动中的仿真成像研究[J]. 光学技术, 2013, 39(1): 28-32. Su Yulu, Cao Fengmei, Wang Fan, et al. The research on the imaging simulation of retina-like sensor in the forward motion [J]. Optical Technique, 2013, 39(1): 28-32. (in Chinese) [33] Kang Wei, Zeng Hansong, Zhao Yi. Insect-Human Hybrid Eye (IHHE): an adaptive optofluidic lens combining the structural characteristics of insect and human eyes [J]. Lab on a Chip, 2014, 14(18): 3594-3602. doi: 10.1039/C4LC00363B [34] David Claveau, Wang Chunyan. Space-variant motion detection for active visual target tracking [J]. Robotics and Autonomous Systems, 2009, 57(1): 11-22. doi: 10.1016/j.robot.2008.04.001 [35] 刘惠中. 空间分辨率可变视觉系统研究[D]. 西安: 西北工业大学, 2005. Liu Huizhong. Research on vision system with variable spatial resolution[D]. Xi'an: Northwestern Polytechnical University, 2005. (in Chinese) [36] 魏新国, 张广军, 江洁. 基于Log-Polar变换的星图识别方法[J]. 光学技术, 2006, 32(5): 678-681. doi: 10.3321/j.issn:1002-1582.2006.05.038 Wei Xinguo, Zhang Guangjun, Jiang Jie. A star identification algorithm base on Log-Polar transform [J]. Optical Technique, 2006, 32(5): 678-681. (in Chinese) doi: 10.3321/j.issn:1002-1582.2006.05.038 [37] 汪洋, 黄小红, 陈曾平. 基于对数极坐标变换的目标识别算法[J]. 雷达科学与技术, 2009, 7(2): 111-114. doi: 10.3969/j.issn.1672-2337.2009.02.007 Wang Yang, Huang Xiaohong, Chen Cengping. A new algorithm for target recognition based on log-polar transform [J]. Radar Science and Technology, 2009, 7(2): 111-114. (in Chinese) doi: 10.3969/j.issn.1672-2337.2009.02.007 [38] Wang Hongsheng, Hao Qun, Cao Jie, et al. Target recognition method with frequency features on retina-like laser detection and range images [J]. Applied Optics, 2019, 58(35): 9532-9539. doi: 10.1364/AO.58.009532 [39] Zhang Kaiyu, Cao Jie, Hao Qun, et al. Modeling and simulations of retina-like three-dimensional computational ghost imaging [J]. IEEE Photonics Journal, 2019, 11: 69007131. [40] 曹杰, 郝群, 宋勇, 等. 基于对数极坐标映射的非均匀透镜阵列设计[J]. 光子学报, 2014, 43: 97-102. Cao Jie, Hao Qun, Song Yong, et al. Designing non-uniform lens array based on log-polar mapping [J]. Acta Photonica Sinica, 2014, 43: 97-102. (in Chinese) [41] Niño-de-Rivera L, Calzada Salas R, Duchen Sánchez G, et al. Visual simulation of retinal images through microstructures [J]. Microelectronic Engineering, 2012, 90: 159-162. doi: 10.1016/j.mee.2011.04.041 [42] Perić Z, Dinčić M, Jovanović A. Optimal log-polar image sampling [J]. Journal of Communications Technology and Electronics, 2009, 54(12): 1397-1402. doi: 10.1134/S1064226909120092 [43] Cheung Brian, Weiss Eric, Olshausen Bruno. Emergence of foveal image sampling from learning to attend in visual scenes[J]. arXiv: 1611.09430, 2017. [44] Bigas M, Cabruja E, Forest J, et al. Review of CMOS image sensors [J]. Microelectronics Journal, 2006, 37(5): 433-451. doi: 10.1016/j.mejo.2005.07.002 [45] Donati S, Martini G, Randone E. Improving photodetector performance by means of microoptics concentrators [J]. Journal of Lightwave Technology, 2011, 29(5): 661-665. doi: 10.1109/JLT.2010.2103302 [46] Daisy S Daivasagaya, Yao Lei, Yung Ka Yi, et al. Contact CMOS imaging of gaseous oxygen sensor array [J]. Sensors and Actuators B: Chemical, 2011, 157(2): 408-416. doi: 10.1016/j.snb.2011.04.074 [47] Huo Yijie, Fesenmaier Christian C, Catrysse Peter B. Microlens performance limits in sub-2μm pixel CMOS image sensors [J]. Optics Express, 2010, 18(6): 5861. doi: 10.1364/OE.18.005861