结合高斯核函数的卷积神经网络跟踪算法
|
摘要
针对视觉跟踪中运动目标鲁棒性跟踪问题,结合高斯核函数和卷积神经网络(CNN),提出一种无需训练的卷积神经网络提取深度特征的视觉跟踪算法.首先,对初始图像进行归一化处理并聚类提取目标信息,结合跟踪过程中目标信息共同作为卷积网络结构中的各阶滤波器;其次,通过高斯核函数来提高卷积运算速度,提取目标简单抽象特征;最后,通过叠加简单层的卷积结果得到目标的深层次表达,并结合粒子滤波跟踪框架实现跟踪.结果表明:简化后的卷积网络结构能够有效地应对低分辨率、目标遮挡与形变等场景,提高复杂背景下的跟踪效率.
In view of the robustness tracking of moving targets in visual tracking,a vision tracking algorithm is proposed in this paper,which combines Gauss kernel function and convolution neural network(CNN),to extract the depth feature of the convolution neural network without training.Firstly,the initial image is normalized and the target information is extracted by clustering,and the target information in the tracking process is combined as the order filter in the convolution network structure.The Gauss kernel function is used to improve the convolution operation speed,extract the simple abstract feature of the target,and then superimpose the convolution results of the simple layer to get the depth of the target.Finally,we combine particle filter tracking framework to achieve tracking.The results show that the simplified convolution network structure can effectively cope with low resolution,target occlusion and deformation and so on,and improve the tracking efficiency in complex background.
In view of the robustness tracking of moving targets in visual tracking,a vision tracking algorithm is proposed in this paper,which combines Gauss kernel function and convolution neural network(CNN),to extract the depth feature of the convolution neural network without training.Firstly,the initial image is normalized and the target information is extracted by clustering,and the target information in the tracking process is combined as the order filter in the convolution network structure.The Gauss kernel function is used to improve the convolution operation speed,extract the simple abstract feature of the target,and then superimpose the convolution results of the simple layer to get the depth of the target.Finally,we combine particle filter tracking framework to achieve tracking.The results show that the simplified convolution network structure can effectively cope with low resolution,target occlusion and deformation and so on,and improve the tracking efficiency in complex background.
引文
[1]杨戈,刘宏.视觉跟踪算法综述[J].智能系统学报,2010,5(2):95-105.DOI:10.3969/j.issn.1673-4785.2010.02.001.
[2]WU Yi,LIM J,YANG M H.Online object tracking:A benchmark[C]∥IEEE Conference on Computer Vision and Pattern Recognition.Portland:IEEE Press,2013:1354-1362.DOI:10.1109/CVPR.2013.312.
[3]BABENKO B,YANG M H,BELONGIE S.Robust object tracking with online multiple instance learning[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2011,33(8):1619-1632.DOI:10.1109/TPAMI.2010.226.
[4]钟必能,潘胜男.选择性搜索和多深度学习模型融合的目标跟踪[J].华侨大学学报(自然科学版),2016,37(2):207-212.DOI:10.11830/ISSN.1000-5013.2016.02.0207.
[5]MEI Xue,LING Haibin.Robust visual tracking using L1minimization[C]∥IEEE International Conference on Computer Vision.[S.l.]:IEEE Computer Society,2009:1436-1443.DOI:10.1109/ICCV.2009.5459292.
[6]ZHANG Kaihua,ZHANG Lei,YANG Minghsuan.Real-time compressive tracking[C]∥Proceedings of the 12th European Conference on Computer Vision-Volume(Ⅲ).Berlin:Springer-Verlag,2012,7574(1):864-877.DOI:10.1007/978-3-642-33712-3_62.
[7]DAVID S,J ROSS,BRUCE A.Visual object tracking using adaptive correlation filters[C]∥IEEE Computer Society Conference on Computer Vision and Pattern Recognition.San Francisco:IEEE Press,2010:2544-2550.DOI:10.1109/CVPR.2010.5539960.
[8]HENRIQUES J F,CASEIRO R,MARTINS P,et al.High-speed tracking with kernelized correlation filters[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2015,37(3):583-596.DOI:10.1109/TPAMI.2014.2345390.
[9]余凯,贾磊,陈雨强,等.深度学习的昨天、今天和明天[J].计算机研究与发展,2013,50(9):1799-1804.
[10]WANG Naiyan,YEUNG D Y.Learning a deep compact image representation for visual tracking[C]∥27th Annual Conference on Neural Information Processing Systems.Nevada:Curran Associates Inc,2013:809-817.
[11]ZHANG Kaihua,LIU Qingshan,WU Yi,et al.Robust tracking via convolutional networks without learning[J].IEEE Transactions on Image Processing,2015,25(4):1779-1792.
[12]NAM H,HAN B.Learning multi-domain convolutional neural networks for visual tracking[C]∥IEEE Conference on Computer Vision and Pattern Recognition.Las Vegas:IEEE Press,2016:4293-4302.DOI:10.1109/CVPR.2016.465.
[13]COMAMICIU D,RAMESH V,MEER P.Real-time trackingof non-rigid objects using mean-shift[C]∥Proceedings of the IEEE Conference Computer Vision and Patter Recogition.Hilton Head Island:IEEE Press,2000:142-149.DOI:10.1109/CVPR.2000.854761.
[14]ISARD M,BLAKE A.Condensation-conditional density propaga-tion for visual tracking[J].International Journal of Computer Vision,1998,29(1):5-28.DOI:10.1023/A:1008078328650.
[15]ROSS D A,LIM J,LIN R S,et al.Incremental learning for robust visual tracking[J].International Journal of Computer Vision,2008,77(1):125-141.DOI:10.1007/s11263-007-0075-7.
[16]WU Y,LIM J,YANG M H.Object tracking benchmark[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2013,9(4):2411-2418.DOI:10.1109/TPAMI.2014.2388226.
[2]WU Yi,LIM J,YANG M H.Online object tracking:A benchmark[C]∥IEEE Conference on Computer Vision and Pattern Recognition.Portland:IEEE Press,2013:1354-1362.DOI:10.1109/CVPR.2013.312.
[3]BABENKO B,YANG M H,BELONGIE S.Robust object tracking with online multiple instance learning[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2011,33(8):1619-1632.DOI:10.1109/TPAMI.2010.226.
[4]钟必能,潘胜男.选择性搜索和多深度学习模型融合的目标跟踪[J].华侨大学学报(自然科学版),2016,37(2):207-212.DOI:10.11830/ISSN.1000-5013.2016.02.0207.
[5]MEI Xue,LING Haibin.Robust visual tracking using L1minimization[C]∥IEEE International Conference on Computer Vision.[S.l.]:IEEE Computer Society,2009:1436-1443.DOI:10.1109/ICCV.2009.5459292.
[6]ZHANG Kaihua,ZHANG Lei,YANG Minghsuan.Real-time compressive tracking[C]∥Proceedings of the 12th European Conference on Computer Vision-Volume(Ⅲ).Berlin:Springer-Verlag,2012,7574(1):864-877.DOI:10.1007/978-3-642-33712-3_62.
[7]DAVID S,J ROSS,BRUCE A.Visual object tracking using adaptive correlation filters[C]∥IEEE Computer Society Conference on Computer Vision and Pattern Recognition.San Francisco:IEEE Press,2010:2544-2550.DOI:10.1109/CVPR.2010.5539960.
[8]HENRIQUES J F,CASEIRO R,MARTINS P,et al.High-speed tracking with kernelized correlation filters[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2015,37(3):583-596.DOI:10.1109/TPAMI.2014.2345390.
[9]余凯,贾磊,陈雨强,等.深度学习的昨天、今天和明天[J].计算机研究与发展,2013,50(9):1799-1804.
[10]WANG Naiyan,YEUNG D Y.Learning a deep compact image representation for visual tracking[C]∥27th Annual Conference on Neural Information Processing Systems.Nevada:Curran Associates Inc,2013:809-817.
[11]ZHANG Kaihua,LIU Qingshan,WU Yi,et al.Robust tracking via convolutional networks without learning[J].IEEE Transactions on Image Processing,2015,25(4):1779-1792.
[12]NAM H,HAN B.Learning multi-domain convolutional neural networks for visual tracking[C]∥IEEE Conference on Computer Vision and Pattern Recognition.Las Vegas:IEEE Press,2016:4293-4302.DOI:10.1109/CVPR.2016.465.
[13]COMAMICIU D,RAMESH V,MEER P.Real-time trackingof non-rigid objects using mean-shift[C]∥Proceedings of the IEEE Conference Computer Vision and Patter Recogition.Hilton Head Island:IEEE Press,2000:142-149.DOI:10.1109/CVPR.2000.854761.
[14]ISARD M,BLAKE A.Condensation-conditional density propaga-tion for visual tracking[J].International Journal of Computer Vision,1998,29(1):5-28.DOI:10.1023/A:1008078328650.
[15]ROSS D A,LIM J,LIN R S,et al.Incremental learning for robust visual tracking[J].International Journal of Computer Vision,2008,77(1):125-141.DOI:10.1007/s11263-007-0075-7.
[16]WU Y,LIM J,YANG M H.Object tracking benchmark[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2013,9(4):2411-2418.DOI:10.1109/TPAMI.2014.2388226.