全景成像技术若干关键问题的研究
|
摘要
全景成像技术连同与其相关的技术,如图像拼接、图像配准等,都属于机器视觉和模式识别的范畴。全景成像系统具有超大视场,大景深、远近皆可清晰成像,以及体积小、重量轻、抗震性能好等显著的优势。在军事领域和民用领域都有十分广阔的应用前景,如军事侦察、安全防盗、公安警戒、医疗仪器、管道监测、地质勘探、深海勘测、空间探索、3-D物体或环境重建、机器人视觉、虚拟环境导航等等,具有很高的研究价值。本文对全景成像技术的几个关键的方面做了深入的研究,主要有以下几项:
1.广泛地研究了国内外的全景成像技术发展状况,对现有的全景图像获取技术、拼接技术和配准技术的几大分类都做了较为深入的了解,并分析了它们各自适用的范围、优点和缺陷,在此基础上,确定了本文的研究方向和目标,是建立一种高分辨率、无畸变的、实时的、低成本的全景图像获取技术。
2.设计并实现了一种以单一固定摄像机和旋转的平面反射镜为核心结构的环形扫描全景图像获取系统。该系统满足研究目标,并且比全景环形透镜(panoramic annular lens,PAL)所获得的柱面全景图像具有更大的竖直方向视场。
3.针对环形扫描全景成像系统的特性,设计并实现了全景图像的拼接技术。利用自主设计的单方向边缘检测,结合灰度投影、互相关算法、最小二乘法拟合直线等方法,确定图像各帧的系统误差,对齐图像。再经过补偿、几何变换、坐标转换等步骤,完成环形图像的拼接。该方法的特点是针对重叠区域极小的待拼接图像,纠正的是旋转量的系统误差。
4.快速电子稳像方法的设计。利用模版图像制作针对平移偏差和旋转偏差的模版向量,利用象限划分、不同坐标系下灰度投影和一维互相关算法,对实时读入的图像做最少的计算,纠正平移和旋转偏差,达到很好配准效果。该方法不但具有速度快、精度较高的优点,而且具有较好的鲁棒性。
Panorama imaging technology, together with relevant technologies such as image mosaic and image registration, belongs to the realm of machine vision and pattern recognition. A panorama imaging system has outstanding advantages of super wide vision field, deep depth of field, focused image for both far and near objects, small bulk, light weight, good aseismatic quality, etc. There is bright martial and civil foreground for its applications, for instance, military reconnaissance, guard against theft, public safety, medical instruments, pipe surveillance, geological exploration, ocean research, space detection, 3-D object or environment rebuilding, robot vision, virtual environment guiding, thus it is of high research value. In this paper, deep study on several key aspects of panorama imaging technology is carried on, which includes the following main issues:
1. Development situations of home and oversea panorama technology is widely studied, including deep understanding on existing image obtaining, mosaic and registration methods and their classifying, as well as analysis of the strong and weak points and fit applying range of each category. Based on above work, the research direction and aim of this paper is determined - to establish a real time panorama technology with high resolution, no aberration, and low cost.
2. A circular scan panorama imaging system is designed and realized; whose cure structure is one fixed camera and a rotating flat reflector. More than meeting the research aim, this system has even larger vertical visual field than panoramic annular lens (PAL).
3. According to features of the circular scan panorama imaging system, the image mosaic technology is also designed and realized. Using mono-directional edge detection, which is self-designed, combined with gray scale projection, inter correlation, curve fitting with least square method,and other arithmetic, the system error of each frame is worked out and these frames are snapped to the same level. Then after steps of compensation, geometric transform and coordinate transform, the circular image mosaic is completed. The uniqueness of this method is that it processes images which have little overlapped sections, and the system error corrected is rotation error.
4. Quick electronic image stabilization method is designed. Making pattern form pattern images vectors to correct plane shift and rotation error, and using quadrant, gray scale projection in different reference frames, and one dimension inter correlation arithmetic, least computation is taken on the newly obtained images to correct the plane shift and rotation error, and good stabilization is achieved. This method is not only of high speed and accuracy, but also being robust.
1.广泛地研究了国内外的全景成像技术发展状况,对现有的全景图像获取技术、拼接技术和配准技术的几大分类都做了较为深入的了解,并分析了它们各自适用的范围、优点和缺陷,在此基础上,确定了本文的研究方向和目标,是建立一种高分辨率、无畸变的、实时的、低成本的全景图像获取技术。
2.设计并实现了一种以单一固定摄像机和旋转的平面反射镜为核心结构的环形扫描全景图像获取系统。该系统满足研究目标,并且比全景环形透镜(panoramic annular lens,PAL)所获得的柱面全景图像具有更大的竖直方向视场。
3.针对环形扫描全景成像系统的特性,设计并实现了全景图像的拼接技术。利用自主设计的单方向边缘检测,结合灰度投影、互相关算法、最小二乘法拟合直线等方法,确定图像各帧的系统误差,对齐图像。再经过补偿、几何变换、坐标转换等步骤,完成环形图像的拼接。该方法的特点是针对重叠区域极小的待拼接图像,纠正的是旋转量的系统误差。
4.快速电子稳像方法的设计。利用模版图像制作针对平移偏差和旋转偏差的模版向量,利用象限划分、不同坐标系下灰度投影和一维互相关算法,对实时读入的图像做最少的计算,纠正平移和旋转偏差,达到很好配准效果。该方法不但具有速度快、精度较高的优点,而且具有较好的鲁棒性。
Panorama imaging technology, together with relevant technologies such as image mosaic and image registration, belongs to the realm of machine vision and pattern recognition. A panorama imaging system has outstanding advantages of super wide vision field, deep depth of field, focused image for both far and near objects, small bulk, light weight, good aseismatic quality, etc. There is bright martial and civil foreground for its applications, for instance, military reconnaissance, guard against theft, public safety, medical instruments, pipe surveillance, geological exploration, ocean research, space detection, 3-D object or environment rebuilding, robot vision, virtual environment guiding, thus it is of high research value. In this paper, deep study on several key aspects of panorama imaging technology is carried on, which includes the following main issues:
1. Development situations of home and oversea panorama technology is widely studied, including deep understanding on existing image obtaining, mosaic and registration methods and their classifying, as well as analysis of the strong and weak points and fit applying range of each category. Based on above work, the research direction and aim of this paper is determined - to establish a real time panorama technology with high resolution, no aberration, and low cost.
2. A circular scan panorama imaging system is designed and realized; whose cure structure is one fixed camera and a rotating flat reflector. More than meeting the research aim, this system has even larger vertical visual field than panoramic annular lens (PAL).
3. According to features of the circular scan panorama imaging system, the image mosaic technology is also designed and realized. Using mono-directional edge detection, which is self-designed, combined with gray scale projection, inter correlation, curve fitting with least square method,and other arithmetic, the system error of each frame is worked out and these frames are snapped to the same level. Then after steps of compensation, geometric transform and coordinate transform, the circular image mosaic is completed. The uniqueness of this method is that it processes images which have little overlapped sections, and the system error corrected is rotation error.
4. Quick electronic image stabilization method is designed. Making pattern form pattern images vectors to correct plane shift and rotation error, and using quadrant, gray scale projection in different reference frames, and one dimension inter correlation arithmetic, least computation is taken on the newly obtained images to correct the plane shift and rotation error, and good stabilization is achieved. This method is not only of high speed and accuracy, but also being robust.
引文
[1]马毅飞,赵文平,全景成像探测技术性能分析与应用,现代防御技术,2006,34(2):63～67
[2]程惠全,姚炜勇,杨国光,朱方明,全景环形成像的转像系统设计,光电工程,2002,29(2):16～19
[3]曾吉勇,苏显渝,双曲面折射全景成像系统,光学学报,2003,23(9):1138～1142
[4]曾吉勇,苏显渝,水平场景无畸变的折射全景成像系统,光学学报,23(5):636～640
[5]曾吉勇,苏显渝,折反射全景成像系统,激光杂志,2004,25(6):62～64
[6]程惠全,姚炜勇,杨国光,朱方明,全景环形成像的转像系统设计,光电工程,2002,29(2):16～19
[7]曾吉勇,苏显渝,柱面场景无畸变的折反射全景成像系统,光电工程,2003,30(1):42～45
[8]曾吉勇,苏显渝,抛物面折反射全景成像系统,光电子激光,2003,14(5):485～488
[9]王伟,陆佩忠,数字图像拼接技术,小型微型计算机系统,2006,27(2):1347～1352
[10]王俊杰,刘家茂,胡运发,于玉,图像拼接技术,计算机科学,2003,30(6):141～144
[11]聂生东,司京玉,医学显微图像自动拼接的方法研究,中国生物医学工程学报,2005,24(2):173～177
[12]王平江,甄恒洲,特征稀少图像的拼接方法,华中科技大学学报(自然科学版),2006,34(8):12～15
[13]梁运行,崔杜武,图像拼接的预处理算法研究,西安理工大学学报,2003,19(4):348～351
[14]Candoeia F M.Jointly registering images in domain and range by piecewise linear comparametric analysis[J].IEEE Transaction Image Processing,2003,12(4):409～419.
[15]Davis J.Mosaics of scenes with moving objects[C].IEEE Computer Society Conference on Computer Vision and Pattern Recognition,1998,354～360.
[16]王明佳,金光,钟平,张炜,实用电子稳像技术原理及方法,光机电信息,2003,2:29～30
[17]葛卫龙雷选华,实现图像稳定的快速算法,红外与激光工程,2004,33(1):43～46
[18]朱娟娟,郭宝龙,冯宗哲,一种基于灰度投影算法的电子稳像方法,光子学报,2005,34(8):1266～1269
[19]屈志毅,王玉珍,钟声伟,图像拼接中特征块的选取,计算机工程,2002,28(9):165～166
[20]樊庆文,王小龙,秦思南,张华,一种改进的变网格模板数字图像快速拼接算法,现代制造工程,2005(2):76～77
[21]方青,王博亮,一种改进的基于比值模板匹配的显微图像拼接,计算机工程,2005,31(24):159～160
[22]马建光,贾云得,一种基于全向摄像机的移动机器人定位方法,北京理工大学学报,2003,23(3):317～321
[23]仵建宁,郭宝龙,冯宗哲,一种基于兴趣点匹配的图像拼接方法,计算机应用,200626(3):610～612
[24]张世阳,王俊杰,胡运发,一种快速全景图像拼接技术,计算机应用与软件,2004,21(3):77～79
[25]平洁,殷润民,一种全景图快速生成算法及实现,微计算机应用,2006,27(1):59～62
[26]林诚凯,李惠,潘金贵,一种全景图生成的改进算法,计算机工程与应用,2004(35):69～71
[27]王仲训,丁挺,丁晓丹,岑伟迪,一种全景图像拼接算法的实现,工程图学学报,2006(2):112～116
[28]侯舒维,郭宝龙,一种图像自动拼接的快速算法,计算机工程,2005,31(15):70～72
[29]逊华燕,周道炳,李生良,一种序列图像的拼接方法,光学精密工程,2000,8(1):35～37
[30]张汗灵,张敏贵,刘庆云,一种序列图像的自动拼接算法,声学与电子工程,2003(2):28～30
[31]李波,一种基于小波和区域的图像拼接方法,电子科技,2005,4:49～52
[32]孙瀚,黄大贵,基于十字形区域搜索法的图像拼接方法,t计量与测试技术,2005,32(1):12～14
[33]石云飞,王永平,李龙济,倪小元,基于模板匹配的柱面全景图拼接优化算法,工程图学学报,2005,1:96～99
[34]李中科,杨晓辉,吴乐南,针对旋转和平移运动的一种图像配准方法,应用科学学报,2005,23(3):282～286
[35]钟平,于前洋,王明佳,金光,提高用于电子稳像的灰度投影算法精度的方法,光电子激光,2003,14(2):182～186
[36]朱娟娟,郭宝龙,冯宗哲,一种基于灰度投影算法的电子稳像方法,光子学报,2005,34(8):1266～1269
[37]王明佳,金光,钟平,张炜,实用电子稳像技术原理及方法,光机电信息,2003,2:29～30
[38]葛卫龙雷选华,实现图像稳定的快速算法,红外与激光工程,2004,33(1):43～46
[39]周渝斌,赵跃进,基于单向投影矢量的数字电子稳像方法,北京理工大学学报,2003,23(4):509～512
[40]聂垣,赵荣椿,康宝生,基于边缘几何特征的图像精确匹配方法,计算机辅助设计与图形学学报,16(12):1668～1673
[41]洪贝,孙继银,图像配准技术研究,战术导弹控制技术,2006,3:109～113
[42]罗钟铉,刘成明,灰度图像匹配的快速算法,计算机辅助设计与图形学学报,2005,17(5):966～970
[43]熊凌,计算机视觉中的图像匹配综述,湖北工业大学学报,2006,21(3):171～173
[44]牛宗标,吴秀清,李滔,基于小面元方法的光学遥感图像快速匹配算法,计算机工程,2006,32(1):203～205
[45]卢炎生,孙钢波,韩根甲,一种基于圆周划分的模板匹配算法,计算机工程与科学,2006,28(1):75～77
[46]梁栋,童强,屈磊,王年,韦穗,一种基于极几何和单应约束的图像匹配算法,系统仿真学报,2006,18(1):44～46
[47]吴均一,朱重光,赵忠明,一种基于小波分析的旋转不变图像快速匹配方法,遥感学报,2002,6(5):339～342
[48]王培容,陈鸿雁,李姣军,一种快速的具有旋转不变性的模板匹配方法,计算机工程与设计,2006,27(14):2667～2672
[49]王旸,冯驰,基于块匹配的电子图像稳定算法,成阳师范学院学报,2006,21(4):36～38
[50](美)R.C冈萨雷斯,数字图像处理(李叔梁译),科学出版社,1983,1～10
[51]David J.Krugliaski, Visual C++技术内幕(潘爱民,王国印译),北京:清华大学出版社,1999
[52]OK系列图像卡硬件安装手册,北京嘉恒中自图像技术有限公司,2004
[53]OK系列图像卡编程手册,北京嘉恒中自图像技术有限公司,2004
[54]徐岸,贾云得,王忠立,基于双曲面镜的全向摄像机视觉实现方法,北京理工大学学报,2004,24(1):82～85
[55]董立羽,卜彦龙,戴斌,电子稳像技术发展评述,信息技术与信息化,2004,6:17～20
[56]张辉,崔杜武,全景图像生成算法的研究与实现,计算机工程,2003,29(6):95～117
[57]王川婴,葛修润,白世伟,数字式全景钻孔摄像系统及应用,岩土力学,2001,22(4):522～525
[58]李义兵,余大昆,林家瑞,邢国泉,实时全景超声成像技术,医疗设备信息,2005,20(10):39～76
[59]廖强,蔡飞,吴存学,基于机器视觉技术的精确定位方法研究及应用,现代制造工程,2005,4:101～103
[60]韩绍坤,赵跃进,刘明奇,电子稳像技术及其发展,光学技术,2001,27(1):71～73
[61]赵红颖,叠宏,熊经武,电子稳像技术概述,光学精密工程,2001,9(4):353～359
[62]罗诗途,王艳玲,张圮,罗飞路,车载图像配准稳定算法中的关键技术,光电工程,2005,32(5):16～20
[2]程惠全,姚炜勇,杨国光,朱方明,全景环形成像的转像系统设计,光电工程,2002,29(2):16～19
[3]曾吉勇,苏显渝,双曲面折射全景成像系统,光学学报,2003,23(9):1138～1142
[4]曾吉勇,苏显渝,水平场景无畸变的折射全景成像系统,光学学报,23(5):636～640
[5]曾吉勇,苏显渝,折反射全景成像系统,激光杂志,2004,25(6):62～64
[6]程惠全,姚炜勇,杨国光,朱方明,全景环形成像的转像系统设计,光电工程,2002,29(2):16～19
[7]曾吉勇,苏显渝,柱面场景无畸变的折反射全景成像系统,光电工程,2003,30(1):42～45
[8]曾吉勇,苏显渝,抛物面折反射全景成像系统,光电子激光,2003,14(5):485～488
[9]王伟,陆佩忠,数字图像拼接技术,小型微型计算机系统,2006,27(2):1347～1352
[10]王俊杰,刘家茂,胡运发,于玉,图像拼接技术,计算机科学,2003,30(6):141～144
[11]聂生东,司京玉,医学显微图像自动拼接的方法研究,中国生物医学工程学报,2005,24(2):173～177
[12]王平江,甄恒洲,特征稀少图像的拼接方法,华中科技大学学报(自然科学版),2006,34(8):12～15
[13]梁运行,崔杜武,图像拼接的预处理算法研究,西安理工大学学报,2003,19(4):348～351
[14]Candoeia F M.Jointly registering images in domain and range by piecewise linear comparametric analysis[J].IEEE Transaction Image Processing,2003,12(4):409～419.
[15]Davis J.Mosaics of scenes with moving objects[C].IEEE Computer Society Conference on Computer Vision and Pattern Recognition,1998,354～360.
[16]王明佳,金光,钟平,张炜,实用电子稳像技术原理及方法,光机电信息,2003,2:29～30
[17]葛卫龙雷选华,实现图像稳定的快速算法,红外与激光工程,2004,33(1):43～46
[18]朱娟娟,郭宝龙,冯宗哲,一种基于灰度投影算法的电子稳像方法,光子学报,2005,34(8):1266～1269
[19]屈志毅,王玉珍,钟声伟,图像拼接中特征块的选取,计算机工程,2002,28(9):165～166
[20]樊庆文,王小龙,秦思南,张华,一种改进的变网格模板数字图像快速拼接算法,现代制造工程,2005(2):76～77
[21]方青,王博亮,一种改进的基于比值模板匹配的显微图像拼接,计算机工程,2005,31(24):159～160
[22]马建光,贾云得,一种基于全向摄像机的移动机器人定位方法,北京理工大学学报,2003,23(3):317～321
[23]仵建宁,郭宝龙,冯宗哲,一种基于兴趣点匹配的图像拼接方法,计算机应用,200626(3):610～612
[24]张世阳,王俊杰,胡运发,一种快速全景图像拼接技术,计算机应用与软件,2004,21(3):77～79
[25]平洁,殷润民,一种全景图快速生成算法及实现,微计算机应用,2006,27(1):59～62
[26]林诚凯,李惠,潘金贵,一种全景图生成的改进算法,计算机工程与应用,2004(35):69～71
[27]王仲训,丁挺,丁晓丹,岑伟迪,一种全景图像拼接算法的实现,工程图学学报,2006(2):112～116
[28]侯舒维,郭宝龙,一种图像自动拼接的快速算法,计算机工程,2005,31(15):70～72
[29]逊华燕,周道炳,李生良,一种序列图像的拼接方法,光学精密工程,2000,8(1):35～37
[30]张汗灵,张敏贵,刘庆云,一种序列图像的自动拼接算法,声学与电子工程,2003(2):28～30
[31]李波,一种基于小波和区域的图像拼接方法,电子科技,2005,4:49～52
[32]孙瀚,黄大贵,基于十字形区域搜索法的图像拼接方法,t计量与测试技术,2005,32(1):12～14
[33]石云飞,王永平,李龙济,倪小元,基于模板匹配的柱面全景图拼接优化算法,工程图学学报,2005,1:96～99
[34]李中科,杨晓辉,吴乐南,针对旋转和平移运动的一种图像配准方法,应用科学学报,2005,23(3):282～286
[35]钟平,于前洋,王明佳,金光,提高用于电子稳像的灰度投影算法精度的方法,光电子激光,2003,14(2):182～186
[36]朱娟娟,郭宝龙,冯宗哲,一种基于灰度投影算法的电子稳像方法,光子学报,2005,34(8):1266～1269
[37]王明佳,金光,钟平,张炜,实用电子稳像技术原理及方法,光机电信息,2003,2:29～30
[38]葛卫龙雷选华,实现图像稳定的快速算法,红外与激光工程,2004,33(1):43～46
[39]周渝斌,赵跃进,基于单向投影矢量的数字电子稳像方法,北京理工大学学报,2003,23(4):509～512
[40]聂垣,赵荣椿,康宝生,基于边缘几何特征的图像精确匹配方法,计算机辅助设计与图形学学报,16(12):1668～1673
[41]洪贝,孙继银,图像配准技术研究,战术导弹控制技术,2006,3:109～113
[42]罗钟铉,刘成明,灰度图像匹配的快速算法,计算机辅助设计与图形学学报,2005,17(5):966～970
[43]熊凌,计算机视觉中的图像匹配综述,湖北工业大学学报,2006,21(3):171～173
[44]牛宗标,吴秀清,李滔,基于小面元方法的光学遥感图像快速匹配算法,计算机工程,2006,32(1):203～205
[45]卢炎生,孙钢波,韩根甲,一种基于圆周划分的模板匹配算法,计算机工程与科学,2006,28(1):75～77
[46]梁栋,童强,屈磊,王年,韦穗,一种基于极几何和单应约束的图像匹配算法,系统仿真学报,2006,18(1):44～46
[47]吴均一,朱重光,赵忠明,一种基于小波分析的旋转不变图像快速匹配方法,遥感学报,2002,6(5):339～342
[48]王培容,陈鸿雁,李姣军,一种快速的具有旋转不变性的模板匹配方法,计算机工程与设计,2006,27(14):2667～2672
[49]王旸,冯驰,基于块匹配的电子图像稳定算法,成阳师范学院学报,2006,21(4):36～38
[50](美)R.C冈萨雷斯,数字图像处理(李叔梁译),科学出版社,1983,1～10
[51]David J.Krugliaski, Visual C++技术内幕(潘爱民,王国印译),北京:清华大学出版社,1999
[52]OK系列图像卡硬件安装手册,北京嘉恒中自图像技术有限公司,2004
[53]OK系列图像卡编程手册,北京嘉恒中自图像技术有限公司,2004
[54]徐岸,贾云得,王忠立,基于双曲面镜的全向摄像机视觉实现方法,北京理工大学学报,2004,24(1):82～85
[55]董立羽,卜彦龙,戴斌,电子稳像技术发展评述,信息技术与信息化,2004,6:17～20
[56]张辉,崔杜武,全景图像生成算法的研究与实现,计算机工程,2003,29(6):95～117
[57]王川婴,葛修润,白世伟,数字式全景钻孔摄像系统及应用,岩土力学,2001,22(4):522～525
[58]李义兵,余大昆,林家瑞,邢国泉,实时全景超声成像技术,医疗设备信息,2005,20(10):39～76
[59]廖强,蔡飞,吴存学,基于机器视觉技术的精确定位方法研究及应用,现代制造工程,2005,4:101～103
[60]韩绍坤,赵跃进,刘明奇,电子稳像技术及其发展,光学技术,2001,27(1):71～73
[61]赵红颖,叠宏,熊经武,电子稳像技术概述,光学精密工程,2001,9(4):353～359
[62]罗诗途,王艳玲,张圮,罗飞路,车载图像配准稳定算法中的关键技术,光电工程,2005,32(5):16～20