Self-tuning hierarchical Kalman-particle filter for efficient target tracking

Xu Chao; Gao Min; Yang Yao

Hierarchical Kalman-particle filter(HKPF) is successfully applied to target tracking with adaption to motion changes. However, it only focuses on the optimization of the target position rather than other affine parameters, resulting in many particles needed to find the optimal state. To achieve fast tracking in complex environment, self-tuning strategy-based hierarchical Kalman-particle filter was proposed to solve the problem. The proposed algorithm marginalized out the linear states in the dynamics to reduce the state dimension, and then found the optimal nonlinear states in a chainlike way with a very small number of particles. The detail process of our algorithm was as follows: first, a local region was estimated by KF; second, self-tuning strategy was used to incrementally generate particles in this region, and an online-learned pose estimator(PE) was introduced to iteratively tune them along the optimal directions according to observations. The comparison among the proposed algorithm and the existing tracking algorithms with real video sequences was implemented, in which the target undergo rapid and erratic motion, or/and dramatic pose change. The results demonstrate that the proposed tracking algorithm can achieve great robustness and very high accuracy with only a very small number of particles.

1. Missile Engineering Department,Ordnance Engineering College,Shijiazhuang 050003,China;

2. Chongqing Optoelectronics Research Institute,Chongqing 400060,China

Received Date: 2014-10-05

Rev Recd Date: 2014-12-15

Publish Date: 2015-06-25

Key words:

Abstract: Hierarchical Kalman-particle filter(HKPF) is successfully applied to target tracking with adaption to motion changes. However, it only focuses on the optimization of the target position rather than other affine parameters, resulting in many particles needed to find the optimal state. To achieve fast tracking in complex environment, self-tuning strategy-based hierarchical Kalman-particle filter was proposed to solve the problem. The proposed algorithm marginalized out the linear states in the dynamics to reduce the state dimension, and then found the optimal nonlinear states in a chainlike way with a very small number of particles. The detail process of our algorithm was as follows: first, a local region was estimated by KF; second, self-tuning strategy was used to incrementally generate particles in this region, and an online-learned pose estimator(PE) was introduced to iteratively tune them along the optimal directions according to observations. The comparison among the proposed algorithm and the existing tracking algorithms with real video sequences was implemented, in which the target undergo rapid and erratic motion, or/and dramatic pose change. The results demonstrate that the proposed tracking algorithm can achieve great robustness and very high accuracy with only a very small number of particles.

HTML

Reference (31)

[1]
[2]	Wang Fasheng, Lu Mingyu, Zhao Qingjie, et al. Particle filtering algorithm [J]. Chinese Journal of Computers, 2014, 37(8): 1679-1694. (in Chinese)
[3]	Thomas V, Ray A K. Fuzzy particle filter for video surveillance[J]. IEEE Transactions on Fuzzy Systems, 2011, 19(5): 937-945.
[4]
[5]	Shana C, Tan T, Wei Y. Real-time hand tracking using a mean shift embedded particle filter[J]. Pattern Recognition, 2007, 40: 1958-1970.
[6]
[7]	Tian Li, Zhou Fugen, Meng Cai. Parallel particle filter object tracking based on embedded multi-core DSP systems[J]. Infrared and Laser Engineering, 2014, 43(7): 2354-2361.
[8]
[9]
[10]	Cao Yang, Zhao Mingfu, Luo Binbin, et al. Airborne platform's tracking algorithm for free space optical [J]. Infrared and Laser Engineering, 2012, 41(11): 3065-3068.
[11]	Cheng H Y, Hwang J N. Adaptive particle sampling and adaptive appearance for multiple video object tracking[J]. Signal Processing, 2009, 89: 1844-1849.
[12]
[13]
[14]	Bimbo A D, Dini F. Particle filter-based visual tracking with a first order dynamic model and uncertainty adaptation[J]. Computer Vision and Image Understanding, 2011, 115: 771-786.
[15]	Kwon J, Lee K M, Park F C. Visual tracking via geometric particle filtering on the affine group with optimal importance functions [C]//IEEE CVPR, 2009: 991-998.
[16]
[17]
[18]	Li P, Zhang T, Pece A. Visual contour tracking based on particle filters[J]. Image Vis Comput, 2003, 21(1): 111-123.
[19]
[20]	Xia Nan, Qiu Tianshuang, Li Jingchun, et al. A nonlinear filtering algorithm combining the kalman filter and the particle filter [J]. Acta Electronic Sinca, 2013, 41(1): 148-152. (in Chinese)
[21]
[22]	Rui Y, Chen Y. Better proposal distributions: object tracking using the unscented particle filter[C]//IEEE Conference on Computer Vision and Pattern Recognition, 2001, 2: 786-793.
[23]	Li Quan, Zhao Xunjie, Peng Qingyan, et al. Windows adaptive particle filter algorithm based on principal component analysis[J]. Infrared and Laser Engineering, 2014, 43(10): 3474-3479. (in Chinese)
[24]
[25]
[26]	Khan Z, Balch T, Dellaert F. A Rao-Blackwellized particle filter for Eigen tracking [C]//IEEE Conference on Computer Vision and Pattern Recognition, 2004: 980-986.
[27]
[28]	Hu Zhentao, Liu Xianxing, Hou Yandong. Real-time marginalized particle filter based on weights consistency optimization[J]. Acta Electronica Sinca, 2014, 42(10): 1970-1976. (in Chinese)
[29]
[30]	Yin S, Na J H, Choi Y J, et al, Hierarchical kalman-particle filter with adaptation to motion changes for object tracking [J]. Computer Vision and Image Understanding, 2011, 115: 885-900.
[31]	Li M, Tan T, Chen W, et al. Efficient object tracking by incremental self-tuning particle filtering on the affine group [J]. IEEE Transactions on Image Processing, 2012, 21(3): 1298-1313.

Self-tuning hierarchical Kalman-particle filter for efficient target tracking

Abstract

References

Proportional views

通讯作者: 陈斌, bchen63@163.com

Article Metrics

Related

Proportional views