基于多模板的深度核相关滤波跟踪

李雪晴; 杨德东; 毛宁; 杨福才

doi:doi:10.3788/yjyxs20173212.0993

液晶与显示, 2017, 32 (12): 993, 网络出版: 2017-12-25

基于多模板的深度核相关滤波跟踪

Depth kernel correlation filtering tracking based on multi-template

论文大纲

李雪晴 ^*杨德东毛宁杨福才

作者单位

河北工业大学控制科学与工程学院, 天津 300130

目标跟踪多模板深度信息重叠率 target tracking multi-template depth information overlap

摘要

针对跟踪过程中出现的遮挡、尺度变化、光照变化等问题, 文章基于多模板提出深度核相关滤波算法。首先, 多模板算法选取最佳滤波参数优化分类器训练样本的能力, 多特征算法利用多种特征优化目标外观模型提高了跟踪过程的鲁棒性; 其次, 利用深度图信息计算跟踪过程中目标重叠率, 判断目标的遮挡情况, 遮挡时重新定义目标搜索区域, 并判断是否重新跟踪目标, 降低遮挡情况下的算法漂移问题; 最后, 根据目标遮挡情况判断是否更新分类器参数和目标外观模型, 提高模板更新的可靠性。利用Princeton数据库测试算法, 成功率和精度分别达到85.1和98.6, 比第二名算法分别提高了7.04%和4.67%。实验从成功率、精确度方面说明基于多模板的深度核相关滤波算法优于传统算法, 有一定研究价值。

Abstract

Aiming at the problems of occlusion, scale transformation and illumination change in the tracking process, a depth kernel correlation filtering algorithm is proposed based on the multi-template. Firstly, the multi-template algorithm selects the best filtering parameters to optimize the ability of the classifier training samples. The various features are used to optimize the target appearance model which improves the robustness of the multi-feature algorithm. Then, in the tracking process, the depth map information is applied to calculate the target overlap rate which is employed to judge whether the tracking target is occluded. When occlusion occurs, the target search area is redefined, and the target is judge whether to track gain. So it reduces the problem of algorithm drift in the case of occlusion. Finally, according to whether or not occlusion occurs, the classifier parameters and target appearance model are determined whether to update. It improves the reliability of template updates. Using the Princeton database test algorithm, the success rate and accuracy is 85.1% and 98.6% respectively, which is 7.04% and 4.67% higher than the second algorithm respectively. Experiments show that the depth kernel correlation filtering algorithm based on multi-template is superior to the traditional algorithm, and has certain research value.

参考文献

[1] 张雷, 王延杰, 孙宏海, 等．采用核相关滤波器的自适应尺度目标跟踪［J］．光学精密工程, 2016, 24(2): 448-459．

ZHANG L, WANG Y J, SUN H H, et al. Adaptive scale object tracking with kernelized correlation filters ［J］. Optics and Precision Engineering, 2016, 24(2): 448-459. (in Chinese)

[2] COMANICIU D, RAMESH V, MEER P. Kernel-based object tracking ［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2003, 25(5): 564-577.

[3] KWON J, LEE K M. Visual tracking decomposition ［C］//Proceedings of 2010 IEEE Conference on Computer Vision and Pattern Recognition. San Francisco, CA: IEEE, 2010: 1269-1276.

[4] KALAL Z, MIKOLAJCZYK K, MATAS J. Tracking-learning-detection ［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012, 34(7): 1409-1422.

[5] 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.

[6] LI Y, ZHU J. A scale adaptive kernel correlation filter tracker with feature integration ［M］//AGAPITO L, BRONSTEIN M, ROTHER C. European Conference on Computer Vision. Cham: Springer, 2014: 254-265.

[7] BERTINETTO L, VALMADRE J, GOLODETZ S, et al. Staple: complementary learners for real-time tracking ［C］//Proceedings of 2016 IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas, NV: IEEE, 2016: 1401-1409.

[8] DANELLJAN M, HGER G, KHAN F S, et al. Discriminative scale space tracking ［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(8): 1561-1575.

[9] CAMPLANI M, HANNUNA S, MIRMEHDI M, et al. Real-time RGB-D tracking with depth scaling kernelised correlation filters and occlusion handling ［C］//XIE X H, JONES M W, TAM G K L. Proceedings of 2015 British Machine Vision Conference. Swansea, UK: BMVA, 2015: 145.1-145.11.

[10] SONG S R, XIAO J X. Tracking revisited using RGBD camera: unified benchmark and baselines ［C］//Proceedings of 2013 IEEE International Conference on Computer Vision. Sydney, NSW: IEEE, 2013: 233-240.

[11] GARCA G M, KLEIN D A, STCKLER J, et al. Adaptive multi-cue 3D tracking of arbitrary objects ［C］//PINZ A, POCK T, BISCHOF H, et al. Joint DAGM (German Association for Pattern Recognition) and OAGM Symposium. Berlin, Heidelberg: Springer, 2012, 7476: 357-366.

[12] BIBI A, GHANEM B. Multi-template scale-adaptive kernelized correlation filters ［C］//Proceedings of 2015 IEEE International Conference on Computer Vision Workshop. Santiago: IEEE, 2015: 613-620.

[13] 张健, 李宏升．基于图论阈值算法的图像分割研究［J］．液晶与显示, 2014, 29(4): 592-597．

ZHANG J, LI H S.Image mosaic research based on wavelet and rough set algorithm ［J］. Chinese Journal of Liquid Crystals and Displays, 2014, 29(4): 592-597. (in Chinese)

[14] 陈恺, 陈芳, 戴敏, 等．基于萤火虫算法的二维熵多阈值快速图像分割［J］．光学精密工程, 2014, 22(2): 517-523．

CHEN K, CHEN F, DAI M,et al. Fast image segmentation with multilevel threshold of two-dimensional entropy based on firefly algorithm ［J］. Optics and Precision Engineering, 2014, 22(2): 517-523. (in Chinese)

李雪晴, 杨德东, 毛宁, 杨福才. 基于多模板的深度核相关滤波跟踪[J]. 液晶与显示, 2017, 32(12): 993. LI Xue-qing, YANG De-dong, MAO Ning, YANG Fu-cai. Depth kernel correlation filtering tracking based on multi-template[J]. Chinese Journal of Liquid Crystals and Displays, 2017, 32(12): 993.

基于多模板的深度核相关滤波跟踪

关于本站 Cookie 的使用提示

全站搜索

基于多模板的深度核相关滤波跟踪

相关论文

相关资讯

关于本站 Cookie 的使用提示

全站搜索