Citation: | Fan Xiangsuo, Xu Zhiyong, Zhang Jianlin. Dim small target tracking based on improved particle filter[J]. Opto-Electronic Engineering, 2018, 45(8): 170569. doi: 10.12086/oee.2018.170569 |
[1] | 苗晓孔, 王春平.改进Sobel算子的单帧红外弱小目标检测[J].光电工程, 2016, 43(12): 119-125. doi: 10.3969/j.issn.1003-501X.2016.12.019 Miao X K, Wang C P. Single frame infra-red (IR) dim small target detection based on improved sobel operator[J]. Opto-Electronics Engineering, 2016, 43(12): 119-125. doi: 10.3969/j.issn.1003-501X.2016.12.019 |
[2] | 王鑫, 唐振民.一种新的复杂背景下红外弱小目标检测方法[J].系统仿真学报, 2009, 21(20): 6568-6572. Wang X, Tang Z. New Method for Infrared Small Target Detection under Complex Background[J]. Journal of System Simulation, 2009, 21(20): 6568-6572. |
[3] | Wang X, Liu L, Tang Z M. Infrared human tracking with improved mean shift algorithm based on multicue fusion[J]. Applied Optics, 2009, 48(21): 4201-4212. doi: 10.1364/AO.48.004201 |
[4] | 王继平, 孙华燕, 章喜.基于Kalman滤波的红外弱小目标检测前跟踪算法[J].装备学院学报, 2012, 23(2): 72-77. Wang J P, Sun H Y, Zhang X. Track-before-detect algorithm for infrared dim target based on kalman filter[J]. Journal of Academy of Equipment, 2012, 23(2): 72-77. |
[5] | Zhan R H, Wan J W. Iterated unscented Kalman filter for passive target tracking[J]. IEEE Transactions on Aerospace and Electronic Systems, 2007, 43(3): 1155-1163. doi: 10.1109/TAES.2007.4383605 |
[6] | 康莉, 谢维信, 黄敬雄.基于unscented粒子滤波的红外弱小目标跟踪[J].系统工程与电子技术, 2007, 29(1): 1-4. Kang L, Xie W X, Huang J X. Tracking of infrared small target based on unscented particle filtering[J]. Systems Engineering and Electronics, 2007, 29(1): 1-4. |
[7] | Han H, Ding Y S, Hao K R, et al. Particle filter for state estimation of jump Markov nonlinear system with application to multi-targets tracking[J]. International Journal of Systems Science, 2013, 44(7): 1333-1343. doi: 10.1080/00207721.2012.737486 |
[8] | Han H, Ding Y S, Hao K R, et al. An evolutionary particle filter with the immune genetic algorithm for intelligent video target tracking[J]. Computers & Mathematics with Applications, 2011, 62(7): 2685-2695. |
[9] | 种衍文, 王泽文, 陈蓉, 等.一种多特征自适应融合的粒子滤波红外目标跟踪方法[J].武汉大学学报·信息科学版, 2016, 41(5): 598-604. Chong Y W, Wang Z W, Chen R, et al. A particle filter infrared target tracking method based on multi-feature adaptive fusion[J]. Geomatics and Information Science of Wuhan University, 2016, 41(5): 598-604. |
[10] | 王鑫, 唐振民.复杂背景下基于改进粒子滤波的红外人体跟踪[J].系统仿真学报, 2010, 22(10): 656-663. Wang X, Tang Z. Infrared human tracking based on improved particle filter under complex background[J]. Journal of System Simulation, 2010, 22(10): 656-663. |
[11] | 王鑫, 唐振民.基于特征融合的粒子滤波在红外小目标跟踪中的应用[J].中国图象图形学报, 2010, 15(1): 91-97. doi: 10.11834/jig.20100115 Wang X, Tang Z. Application of particle filter based on feature fusion in small IR target tracking[J]. Journal of Image and Graphics, 2010, 15(1): 91-97. doi: 10.11834/jig.20100115 |
[12] | Wang W G, Li C M, Shi J N. A robust infrared dim target detection method based on template filtering and saliency extraction[J]. Infrared Physics & Technology, 2015, 73: 19-28. |
[13] | Wang X, Shen S Q, Ning C, et al. A sparse representation-based method for infrared dim target detection under sea-sky background[J]. Infrared Physics & Technology, 2015, 71: 347-355. |
[14] | Li Z Z, Chen J, Hou Q, et al. Sparse representation for infrared dim target detection via a discriminative over-complete dictionary learned online[J]. Sensors, 2014, 14(6): 9451-9470. doi: 10.3390/s140609451 |
[15] | Wang L J, Ouyang W L, Wang X G, et al. Visual tracking with fully convolutional networks[C]// Proceedings of 2015 IEEE International Conference on Computer Vision. Santiago, Chile: IEEE, 2015. |
[16] | Ma C, Huang J B, Yang X K, et al. Hierarchical convolutional features for visual tracking[C]// Proceedings of 2015 IEEE International Conference on Computer Vision. Santiago, Chile: IEEE, 2015. |
[17] | Yang X, Sun H, Fu K, et al. Automatic ship detection in remote sensing images from google earth of complex scenes based on multiscale rotation dense feature pyramid networks[J]. Remote Sensing, 2018, 10(1): 132-139. doi: 10.3390/rs10010132 |
[18] |
王鑫. 复杂背景下红外目标检测与跟踪算法研究[D]. 南京: 南京理工大学, 2010.
Wang X. Infrared target detection and tracking algorithms under complex background[D]. Nanjing: Nanjing University of Science and Technology, 2010. |
[19] | Fan X, Xu Z, Zhang J, et al. Dim small targets detection based on self-adaptive caliber temporal-spatial filtering[J]. Infrared Physics & Technology, 2017, 85: 465-477. |
[20] | 张强, 蔡敬菊, 张启衡, 等.基于局部极大值的红外弱小目标分割方法[J].红外技术, 2011, 33(1): 41-44. Zhang Q, Cai J J, Zhang Q H, et al. Small dim infrared targets segmentation method based on local maximum[J]. Infrared Technology, 2011, 33(1): 41-44. |
[21] | Shan C F, Wei Y C, Tan T N, et al. Real time hand tracking by combining particle filtering and mean shift[C]// Proceedings of the 6th IEEE International Conference on Automatic Face and Gesture Recognition. Seoul: IEEE, 2004. |
Overview: At present, tracking algorithms of infrared weak and small targets mainly include single-frame detection algorithm, Meanshift algorithm, Kalman filtering, extended Kalman filter, and unscented Kalman filter. Single-frame detection algorithms achieve certain detection results in a specified scenario. However, the algorithms do not make the most of the motion information of the target in different frames, with certain limitations. Meanshift algorithm is easy to lead to tracking failure under the condition of the dynamic changing background or the occluded target. The Kalman filtering algorithm is only applicable to the linear Gaussian units. Extended Kalman filter can effectively solve the target tracking problem of nonlinear systems, but it is hard to solve the filter divergence problem. Untracked Kalman filter algorithm in order to solve the filter divergence problem, but the calculation error is relatively large with the method. Because the particle filter can greatly solve the problems in the nonlinear and non-Gaussian scenes, it has been developed as a significant research direction of weak and small target tracking algorithms. However, for similar background or strong noise interference, the traditional particle filter is prone to produce tracking loss in the following two conditions: 1) The target size is small. Relying only on the gray or shape characteristics is easy to lead to unstable. 2) In low SNR < 3 dB scenarios, it is prone to produce tracking loss due to high noise interference. In order to solve the problem of tracking loss that only relying on the single feature in low signal-to-noise ratio scene, some researchers have proposed a particle tracking filtering based on multi-feature fusion.
As to solve the problem of dim small target tracking in low SNR scenes, an improved particle filter tracking method is proposed. This paper firstly obtains the gray feature by spatial position weighting method, and combines the neighborhood motion model and the gray probability graph to get the motion features of dim small target. Then construct the joint observation model of gray and motion features to calculate the particle weights. In the process of tracking, the gray distribution of the target is not stable, and the strategy of adaptively updating the gray template of reference target is added. Finally, the sequence image is used to prove the tracking effect of dim small target. Experiments show that compared with the traditional particle filter algorithm, the proposed method greatly enhanced the tracking ability of dim small target in low SNR scenes.
Target movement model
Flow chart
Relationship between SNR gain and cumulative frame length
Probability distribution map of motion. (a) The 1st frame; (b) The motion model graph; (c) The gray probability graph; (d) The motion probability graph
Tracking results of particle filter. (a) The 5th frame; (b) The 55th frame; (c) The 61th frame; (d) The 96th frame
Tracking results of the proposed method. (a) The 5th frame; (b) The 55th frame; (c) The 61th frame; (d) The 96th frame
Tracking error curve of scene 1
Tracking results of particle filter. (a) The 25th frame; (b) The 48th frame; (c) The 80th frame; (d) The 101th frame
Tracking results of the proposed method. (a) The 25th frame; (b) The 48th frame; (c) The 80th frame; (d) The 101th frame
Tracking error curve of scene 2
Tracking results of particle filter. (a) The 234th frame; (b) The 239th frame; (c) The 250th frame; (d) The 277th frame
Tracking results of the proposed method. (a) The 234th frame; (b) The 239th frame; (c) The 250th frame; (d) The 277th frame
Tracking error curve of scene 3