利用无监督聚类实现深度图像的遮挡边界检测
|
摘要
提出一种利用无监督聚类思想检测深度图像中遮挡边界的方法.首先根据遮挡边界点与其邻域点的空间及深度信息提出了加权最长线段特征并定义了其计算方法.其次,结合遮挡边界点与其邻域点的特征值分布情况提出了一种非线性归一化方法归一化遮挡相关特征.然后,以像素点为单位将各遮挡相关特征组成联合特征向量输入到无监督聚类分类器中,判断待测像素点是否为遮挡边界点.最后,将遮挡边界点可视化得到深度图像的遮挡边界.实验结果表明,无需标记样本的所提方法对深度图像中目标物体的遮挡检测效果同已有的基于监督学习方法的检测效果相当.
An occlusion boundary detection approach is proposed for depth image utilizing unsupervised clustering. Firstly,a feature named weighted longest line segment and its computing method are proposed based on the spatial and depth information between occlusion boundary points and their neighbor points. Secondly,a novel nonlinear normalization method is proposed to normalization the occlusion related features according to the feature distribution of occlusion boundary points and their neighbor points. Thirdly,each pixel's combined feature vector,which is composed of occlusion related features,is inputted into the unsupervised clustering classifier to judge whether the pixel is occlusion boundary point. Finally,the occlusion boundary of depth image can be obtained by visualizing the occlusion boundary points. Experimental results show the detection effect of proposed approach which does not need any labeled samples is equivalent to that of the existing supervised-based learning method.
An occlusion boundary detection approach is proposed for depth image utilizing unsupervised clustering. Firstly,a feature named weighted longest line segment and its computing method are proposed based on the spatial and depth information between occlusion boundary points and their neighbor points. Secondly,a novel nonlinear normalization method is proposed to normalization the occlusion related features according to the feature distribution of occlusion boundary points and their neighbor points. Thirdly,each pixel's combined feature vector,which is composed of occlusion related features,is inputted into the unsupervised clustering classifier to judge whether the pixel is occlusion boundary point. Finally,the occlusion boundary of depth image can be obtained by visualizing the occlusion boundary points. Experimental results show the detection effect of proposed approach which does not need any labeled samples is equivalent to that of the existing supervised-based learning method.
引文
[1]Stein A,Hebert M.Occlusion boundaries from motion:low-level detection and mid-level reasoning[J].International Journal of Computer Vision,2009,82(3):325-357.
[2]Schmaltz C,Rosenhahn B,Brox T,et al.Region-based pose tracking w ith occlusions using 3D models[J].M achine Vision and Applications,2012,23(3):557-577.
[3]Lee S J,Park K R,Kim J.A Sf M-based 3D face reconstruction method robust to self-occlusion by using a shape conversion matrix[J].Pattern Recognition,2011,44(7):1470-1486.
[4]Alvarez L,Deriche R,Papadopoulo T.Symmetrical dense optical flow estimation w ith occlusions detection[J].International Journal of Computer Vision,2007,75(3):371-385.
[5]He X,Yuille A.Occlusion boundary detection using pseudo-depth[C].Proceedings of the European Conference on Computer Vision,Heidelberg,Germany:Springer Verlag,2010:539-552.
[6]Humayun A,Aodha O M,Brostow G J.Learning to find occlusion regions[C].Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition,Piscataw ay,USA:IEEE,2011:2161-2168.
[7]Chen D,Yuan Z,Zhang G,et al.Detecting occlusion boundaries via saliency netw ork[C].Proceedings of International Conference on Pattern Recognition,Tsukuba,Piscataw ay,USA:IEEE,2012:2569-2572.
[8]Zhang Shi-hui,Zhang Yu-jie,Kong Ling-fu.Self-occlusion detection approach based on depth image[J].Journal of Chinese Computer Systems,2010,31(5):964-968.
[9]Zhang S,Gao F,Kong L.A self-occlusion detection approach based on range image of vision object[J].Innovative Computing,Information and Control Express Letters,2011,5(6):2041-2046.
[10]Zhang Shi-hui,Liu Jian-xin,Kong Ling-fu.Using random forest for occlusion detection based on depth image[J].Acta Optica Sinica,2014,34(9):1-12.
[11]Zhang Shi-hui,Pang Yun-chong.Occlusion boundary detection method for depth image based on ensemble learning[J].Acta M etrologica Sinica,2014,35(6):569-573.
[12]Zhang D,Han J,Cheng G,et al.Weakly supervisedlearning for target detection in remote sensing images[J].Geoscience and Remote Sensing Letters,IEEE,2015,12(4):701-705.
[13]Han J,Zhang D,Cheng G,et al.Object detection in optical remote sensing images based on w eakly supervised learning and high-level feature learning[J].IEEE Transactions on Geoscience and Remote Sensing,2015,53(6):3325-3337.
[14]Geng X,Zhan D C,Zhou Z H.Supervised nonlinear dimensionality reduction for visualization and classification[J].IEEE Transactions on Systems,M an,and Cybernetics,Part B:Cybernetics,2005,35(6):1098-1107.
[15]Breiman L.Random forests[J].Machine Learning,2001,45(1):5-32.
[16]Hetzel G,Leibe B,Levi P,et al.3D object recognition from range images using local feature histrograms[C].Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition,Kauai,HI,USA,2001.II394-II399.
[8]张世辉,张煜婕,孔令富.一种基于深度图像的自遮挡检测方法[J].小型微型计算机系统,2010,31(5):964-968.
[10]张世辉,刘建新,孔令富.基于深度图像利用随机森林实现遮挡检测[J].光学学报,2014,34(9):1-12.
[11]张世辉,庞云冲.基于集成学习思想的深度图像遮挡边界检测方法[J].计量学报,2014,35(6):569-573.
[2]Schmaltz C,Rosenhahn B,Brox T,et al.Region-based pose tracking w ith occlusions using 3D models[J].M achine Vision and Applications,2012,23(3):557-577.
[3]Lee S J,Park K R,Kim J.A Sf M-based 3D face reconstruction method robust to self-occlusion by using a shape conversion matrix[J].Pattern Recognition,2011,44(7):1470-1486.
[4]Alvarez L,Deriche R,Papadopoulo T.Symmetrical dense optical flow estimation w ith occlusions detection[J].International Journal of Computer Vision,2007,75(3):371-385.
[5]He X,Yuille A.Occlusion boundary detection using pseudo-depth[C].Proceedings of the European Conference on Computer Vision,Heidelberg,Germany:Springer Verlag,2010:539-552.
[6]Humayun A,Aodha O M,Brostow G J.Learning to find occlusion regions[C].Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition,Piscataw ay,USA:IEEE,2011:2161-2168.
[7]Chen D,Yuan Z,Zhang G,et al.Detecting occlusion boundaries via saliency netw ork[C].Proceedings of International Conference on Pattern Recognition,Tsukuba,Piscataw ay,USA:IEEE,2012:2569-2572.
[8]Zhang Shi-hui,Zhang Yu-jie,Kong Ling-fu.Self-occlusion detection approach based on depth image[J].Journal of Chinese Computer Systems,2010,31(5):964-968.
[9]Zhang S,Gao F,Kong L.A self-occlusion detection approach based on range image of vision object[J].Innovative Computing,Information and Control Express Letters,2011,5(6):2041-2046.
[10]Zhang Shi-hui,Liu Jian-xin,Kong Ling-fu.Using random forest for occlusion detection based on depth image[J].Acta Optica Sinica,2014,34(9):1-12.
[11]Zhang Shi-hui,Pang Yun-chong.Occlusion boundary detection method for depth image based on ensemble learning[J].Acta M etrologica Sinica,2014,35(6):569-573.
[12]Zhang D,Han J,Cheng G,et al.Weakly supervisedlearning for target detection in remote sensing images[J].Geoscience and Remote Sensing Letters,IEEE,2015,12(4):701-705.
[13]Han J,Zhang D,Cheng G,et al.Object detection in optical remote sensing images based on w eakly supervised learning and high-level feature learning[J].IEEE Transactions on Geoscience and Remote Sensing,2015,53(6):3325-3337.
[14]Geng X,Zhan D C,Zhou Z H.Supervised nonlinear dimensionality reduction for visualization and classification[J].IEEE Transactions on Systems,M an,and Cybernetics,Part B:Cybernetics,2005,35(6):1098-1107.
[15]Breiman L.Random forests[J].Machine Learning,2001,45(1):5-32.
[16]Hetzel G,Leibe B,Levi P,et al.3D object recognition from range images using local feature histrograms[C].Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition,Kauai,HI,USA,2001.II394-II399.
[8]张世辉,张煜婕,孔令富.一种基于深度图像的自遮挡检测方法[J].小型微型计算机系统,2010,31(5):964-968.
[10]张世辉,刘建新,孔令富.基于深度图像利用随机森林实现遮挡检测[J].光学学报,2014,34(9):1-12.
[11]张世辉,庞云冲.基于集成学习思想的深度图像遮挡边界检测方法[J].计量学报,2014,35(6):569-573.