基于MSCR和ASIFT的鱼眼图像立体匹配研究
|
摘要
鱼眼镜头在具有大视场优点的同时,存在成像失真大的缺陷,特别在边缘区域更为严重,因而在光轴正交或大视场条件下的立体视觉系统中,难以实现对图像的特征匹配。常规立体视觉匹配方法多基于小孔成像模型进行构建,直接用于鱼眼镜头图像的匹配存在较大误差,会影响图像间匹配的准确度。针对此问题,建立了一种具有仿射不变性的联合匹配算法。先提取鱼眼镜头图像的彩色图像最大稳定值区域(MSCR)特征区域,接着采用仿射尺度不变特征变换(ASIFT)算子对该区域进行逐一描述,从而获取鱼眼图像间的最佳匹配。通过试验验证,该方法具有稳定一致性,能够完成未经失真校正鱼眼镜头图像的立体匹配。
Fisheye lens has advantage of large field of view,but it also has large imaging distortion at the same time,especially more seriously in the edge region,so it is difficult for image feature matching in such a large field of view or orthogonal optical axis condition of stereo vision system. Conventional stereo matching methods are based on pinhole imaging model,which has great errors in direct use of fisheye lens image matching,thus the image matching accuracy is affected. To solve this problem,a combined matching algorithm with affine invariant solution is setup. Firstly,MSCR feature regions of the fisheye image are extracted; then ASIFT algorithm is adopted to describe the characteristic regions of MSCR one by one,so the optimal matching between fisheye images is obtained. The experiments verify the stability consistency of this method; it can complete the stereo matching of fisheye images without distortion correction.
Fisheye lens has advantage of large field of view,but it also has large imaging distortion at the same time,especially more seriously in the edge region,so it is difficult for image feature matching in such a large field of view or orthogonal optical axis condition of stereo vision system. Conventional stereo matching methods are based on pinhole imaging model,which has great errors in direct use of fisheye lens image matching,thus the image matching accuracy is affected. To solve this problem,a combined matching algorithm with affine invariant solution is setup. Firstly,MSCR feature regions of the fisheye image are extracted; then ASIFT algorithm is adopted to describe the characteristic regions of MSCR one by one,so the optimal matching between fisheye images is obtained. The experiments verify the stability consistency of this method; it can complete the stereo matching of fisheye images without distortion correction.
引文
[1]JAN H,TOMAS P.Stereographic rectification of omnidirectional stereo pairs[C]//2009 IEEE Conference on Computer Vision and Pattern Recognition,2009:1414-1421.
[2]项志宇.折反射全向相机镜面位资的自标定方法[J].光学学报,2011,32(4):126-132.
[3]朱均超,葛磊,韩芳芳,等.基于分离参数标定的鱼眼镜头成像模型研究[J].传感技术学报,2013,26(7):975-980.
[4]冯为嘉,张宝峰,曹作良.基于鱼眼镜头的全方位视觉参数标定与畸变矫正[J].天津大学学报,2011,44(5):417-424.
[5]于红绯,刘威,赵宏.基于子块运动补偿的运动目标检测[J].电子学报,2017,45(1):173-180.
[6]徐敏,张红英.吴亚东.基于场景深度的雾天图像能见度检测算法[J].自动化仪表,2017,38(9):89-95.
[7]过志强,殷国亮,肖文皓,等.视觉引导的机器人关节空间动态轨迹规划[J].自动化仪表,2015,36(3):77-80.
[8]张旭苹,汪家其,张益昕,等.大尺度三维几何尺寸立体视觉测量系统实现[J].光学学报,2012,32(3):148-155.
[9]姜宏志,赵慧洁,梁宵月,等.基于极线校正的快速相位立体匹配[J].光学精密工程,2011,19(10):2520-2525.
[10]宋阳,周亚丽,张奇志.鱼眼镜头径向畸变的校正方法[J].仪器仪表学报,2017,38(4):1014-1023.
[11]祝海江,徐晓波,周靖林.基于球面透视投影旋转矩阵的鱼眼图像匹配[J].光学学报,2013,33(2):1-8.
[12]FORSSN P E,LOWE D G.Shape descriptors for maximally stable extremal regions[C]//IEEE 11th International Conference on Computer Vision.IEEE,2007:1-8.
[13]丁文锐,康传波,李红光,等.基于MSER的无人机图像建筑区域提取[J].北京航空航天大学学报,2015,41(3):383-390.
[14]柳涛.多通道图像MSER局部不变特征提取算法研究[D].长沙:国防科学技术大学,2010.
[15]高士林,吉立新,李绍梅,等.基于MSER拟合椭圆的快速场景文本定位算法[J].计算机工程与设计,2015,36(3):693-720.
[16]廉蔺,王海涛,李国辉,等.基于MSER的红外与可见光图像关联特征提取算法[J].电子与信息学报,2011,33(7):1625-1631.
[17]杨杰,刘小军,孙坚伟,等.基于SIFT的图像配准方法[J].红外与激光工程,2008,37(1):156-160.
[18]杨小辉,王敏.基于ASIFT的无缝图像接方法[J].计算机工程,2013,39(2):241-249.
[19]傅卫平,秦川,刘佳,等.基于SIFT算法的图像目标匹配与定位[J].仪器仪表学报,2011,32(1):163-169.
[20]刘佳,傅卫平,王雯,等.基于改进SIFT算法的图像匹配[J].仪器仪表学报,2013,34(5):1107-1112.
[21]薛艳华,何小海,雷翔,等.基于改进ASIFT算法的遥感图像匹配[J].四川大学学报,2013,50(5):1000-1005.
[2]项志宇.折反射全向相机镜面位资的自标定方法[J].光学学报,2011,32(4):126-132.
[3]朱均超,葛磊,韩芳芳,等.基于分离参数标定的鱼眼镜头成像模型研究[J].传感技术学报,2013,26(7):975-980.
[4]冯为嘉,张宝峰,曹作良.基于鱼眼镜头的全方位视觉参数标定与畸变矫正[J].天津大学学报,2011,44(5):417-424.
[5]于红绯,刘威,赵宏.基于子块运动补偿的运动目标检测[J].电子学报,2017,45(1):173-180.
[6]徐敏,张红英.吴亚东.基于场景深度的雾天图像能见度检测算法[J].自动化仪表,2017,38(9):89-95.
[7]过志强,殷国亮,肖文皓,等.视觉引导的机器人关节空间动态轨迹规划[J].自动化仪表,2015,36(3):77-80.
[8]张旭苹,汪家其,张益昕,等.大尺度三维几何尺寸立体视觉测量系统实现[J].光学学报,2012,32(3):148-155.
[9]姜宏志,赵慧洁,梁宵月,等.基于极线校正的快速相位立体匹配[J].光学精密工程,2011,19(10):2520-2525.
[10]宋阳,周亚丽,张奇志.鱼眼镜头径向畸变的校正方法[J].仪器仪表学报,2017,38(4):1014-1023.
[11]祝海江,徐晓波,周靖林.基于球面透视投影旋转矩阵的鱼眼图像匹配[J].光学学报,2013,33(2):1-8.
[12]FORSSN P E,LOWE D G.Shape descriptors for maximally stable extremal regions[C]//IEEE 11th International Conference on Computer Vision.IEEE,2007:1-8.
[13]丁文锐,康传波,李红光,等.基于MSER的无人机图像建筑区域提取[J].北京航空航天大学学报,2015,41(3):383-390.
[14]柳涛.多通道图像MSER局部不变特征提取算法研究[D].长沙:国防科学技术大学,2010.
[15]高士林,吉立新,李绍梅,等.基于MSER拟合椭圆的快速场景文本定位算法[J].计算机工程与设计,2015,36(3):693-720.
[16]廉蔺,王海涛,李国辉,等.基于MSER的红外与可见光图像关联特征提取算法[J].电子与信息学报,2011,33(7):1625-1631.
[17]杨杰,刘小军,孙坚伟,等.基于SIFT的图像配准方法[J].红外与激光工程,2008,37(1):156-160.
[18]杨小辉,王敏.基于ASIFT的无缝图像接方法[J].计算机工程,2013,39(2):241-249.
[19]傅卫平,秦川,刘佳,等.基于SIFT算法的图像目标匹配与定位[J].仪器仪表学报,2011,32(1):163-169.
[20]刘佳,傅卫平,王雯,等.基于改进SIFT算法的图像匹配[J].仪器仪表学报,2013,34(5):1107-1112.
[21]薛艳华,何小海,雷翔,等.基于改进ASIFT算法的遥感图像匹配[J].四川大学学报,2013,50(5):1000-1005.