摘要
光电成像跟踪任务过程中，不同目标呈现出特征多样的特点。即使是同一类目标，其特性在不同阶段、不同背景下也会表现出极大差异。随着目标与跟踪设备距离的变化，其成像会分别表现为扩展目标、面目标、小目标等多种特性。在跟踪过程中，目标甚至还可能因为非均匀光照或发热、分离、熄火等情况而发生成像特征的突变。这些动态、多变的目标特性给成像跟踪算法的适应能力带来了严峻考验，影响了跟踪的稳定性。
针对上述特征动态变化目标的稳定跟踪问题，首先对特定特性目标的可靠提取与稳定跟踪的单模算法进行了研究：针对扩展目标稳定跟踪点提取问题，提出了一种新的基于弦弧比滤波的轮廓平滑算法以及一种基于头部最小内切圆的目标前沿跟踪点定位算法，大幅降低了前沿跟踪的提取点抖动；采用对比度跟踪算法实现了简单背景面目标的稳定跟踪；提出了一种基于二值模板膨胀掩模与改进加权更新的自适应模板更新算法，提高了模板更新的质量并增强了复杂背景下相关跟踪的鲁棒性；研究了小目标检测与稳定跟踪的方法，采用基于灰度形态学与高通滤波结合的方法实现了小目标的可靠检测，通过多特征航迹关联与去背景的质心提取方法，提高了小目标跟踪的稳定性。
其次，针对上述单模算法难以适应目标所有动态变化情况、难以实现目标全程稳定跟踪的问题，研究并提出了粗精两级并行跟踪的框架，提高了目标跟踪的可靠性与稳定性：采用全视场多算法并行处理方法，实现了多目标粗跟踪；为解决不同阶段与不同背景下主目标的稳定跟踪问题，提出了分段式对比度跟踪与相关跟踪并行的双模跟踪方式，大幅提升了系统的稳定跟踪能力以及对目标特征突变情况的适应能力。通过粗精两级并行跟踪的有机结合，实现了一个高智能化、强鲁棒性的成像目标自动检测与跟踪系统。
在系统具备稳定跟踪能力的基础上，研究了基于多特征模糊融合技术的目标航迹事件自动识别方法，实现了目标各种典型航迹事件及其发生时刻的可靠检测。为设备进行场景推理、智能决策、快速评估奠定了基础，提高了系统的智能性与实际应用价值。
最后，为了提高系统实时性以降低动态延迟、确保系统的处理带宽，从图像处理算法的特性入手，讨论了高效硬件平台应当具备的特点以及粗精两级并行跟踪软件向并行硬件平台的映射方法。在此基础上，研究了实时嵌入式软件优化的方法，通过对软件模块实例的实际优化方法的讨论，得出了要在既定硬件平台上实现高效的算法，需要从算法级与代码级进行优化的结论。通过高效的软件优化，进一步提高了系统实时性，确保了系统稳定跟踪的实现。
In the task of photoelectric image tracking, different kinds of targets exhibitdiversity of characteristics. Even the same target, it represents different characteristicsin different tracking phase or background. With the distance increasing between thetarget and the tracking system, the target imaging will exhibit extended target, masstarget and small target successively. The characteristics of the target may enduresudden variation due to asymmetric illumination, splitting and flameout in the courseof the tracking. These dynamic and diverse characteristics of the target influence theadaptabilities of the tracking algorithms and introduce instability into the trackingsystem severely.
For tracking the target with changing characteristics stably, the robustsingle-mode detection and tracking algorithm for special kind of target wereresearched. For the problem of stable tracking point extraction for infrared extendedtarget, a novel approach is proposed to reduce the tracking jitter. A new contoursmoothing method based on the chord-arc ratio filtering is introduced to obtain apreliminary extraction point with lower jitters. Then a novel fine tracking pointextraction method based on the minimal inscribed circle is presented. The targetextraction based on the contrast segmentation method is used for the target tracking inthe simple background. For target stable tracking in the complex background, a noveladaptive template update method is proposed which is based on the binary templatedialation and weighting updation. A method based on the gray mophology and highpass filtering for detection small target is presented. Through the trajectory associationbased on the multi-feature and the method for extracting the centroid with backgroundremoved, the stability of the small target tracking is improved.
Due to the single-mode algorithm can not suit all the variation situations of thetarget and it is difficult to track the target in all the stages, the coarse-fine paralleltracking frame is proposed to improve the tracking stability and reliability. Themulti-target coarse tracking is realized based on the multi-algorithm parallel methodin the whole view of the sight. For main target stable tracking in different phase anddifferent backgound, the double-mode tracking method based on the contrast and correlation is used. Then the capability of the system for stable tracking and theadaptability for variation of the target characteristics is improved. At last, anintelligent and robust system for auto target detection and tracking is realized.
The method for target trajectory event auto recognition based on themulti-feature fuzzy fusion technology is researched, and the robust detection for somerepresentative target trajectory event and its occurrence time is realized. It can helpthe system to make decision, situation reasoning and make fast evaluation. So itimproves the intelligence and the value of the system.
In order to improve the real-time performance of the system, reduce its dynamicdelay and ensure its process bandwith, the characteristics which should be included inan efficient hardware platform were discussed, based on the analyse for the featuresof the image process algorithms. The method for mapping the coarse-fine paralleltracking software onto the parallel hardware platform and the method for optimizingthe embeded software is researched. It concludes that the software must be optimizedbased on both algorithm level and code level for achieving real-time performance fora given hardware platform. Through efficient software optimization, the real-timeperformance of the image-based system is improved and the stable tracking isrealized.
引文
[1]马佳光.捕获跟踪与瞄准系统的基本技术问题[J].光学工程,1989,(3):1-42.
[2]D.Marr. Vision-A computational investigation into the human representation and processing ofvisual information[M]. San Francisco: CA: Freeman,1982.
[3]Andrew.P.Witkin. Scale-space filtering[M]. In Proc.Int. Joint Conf. Artif. Intell, Karlstuhe, WestGermany,1983:1019-1021.
[4]Koenderink.J.J. The structure of images[J]. Biological Cybernetics,1984(50):363-370.
[5]T.Lindeberg. Scale-space theory: A basic tool for analyzing structures at different scales[J].Journal of Applied Statistics,1994,21(2):223-261.
[6]江和平.红外序列图像匹配跟踪技术研究[D].长沙:国防科学技术大学2006.
[7]张焱.地面背景下成像目标跟踪技术研究[D].长沙:国防科学技术大学2008.
[8]张小虎.靶场图像运动目标检测与跟踪定位技术研究[D].长沙:国防科学技术大学2006.
[9]杨卫平,沈振康,红外图像序列小目标检测预处理技术[J].红外与激光工程,1998,27(1):23-28.
[10]黄宗福,王卫华,韩建涛等.一种天文光电图像序列弱小目标实时检测算法[J].信号处理,2010,26(9):1379-1384.
[11]韩建涛,安成锦,黄宗福等.一种基于直线检测的缓动弱小目标检测方法[J].信号处理，2009,25(8A):229-232.
[12]聂洪山.强起伏条件下红外小目标检测算法研究[D].长沙：国防科学技术大学,2004.
[13]张天序.成像自动目标识别[M].湖北武汉:湖北科学技术出版社,2003.
[14]楚灜.智能视频监控中的多特征融合问题研究[D].武汉:华中科技大学,2008.
[15]阎青.复杂背景下目标搜索跟踪方法研究及DSP实现[D].武汉:华中科技大学,2005.
[16]鲜海滢,李晓峰,李在铭.基于梯度相关性的红外微弱目标检测[J].光电子激光.2008,19(9):1214-1219.
[17]魏波.点时空约束图像目标跟踪理论与实时实现技术研究[D].成都:电子科技大学,2002.
[18]艾斯卡尔,李在铭.红外图像序列中微弱点状动目标的非参数检测[J].仪器仪表学报,2002,23(3)增刊:.650-651,659.
[19]Zhenming Peng, Qiheng Zhang, Anquan Guan. Extended target tracking using projection curvesand matching pel count[J]. Opt. Eng.2007,46(6):1-6.
[20]侯志强,韩崇昭.视觉跟踪技术综述[J].自动化学报,2006,32(4):604-617.
[21] Nguyen H T,et al. Occlusion robust adaptive template tracking[A]. Proc. IEEE Int. Conf. onComputer Vision[C]. Vancouver,BC:IEEE,2001.678-683.
[22]潘吉彦,胡波,张建秋.抑制模板漂移的目标跟踪算法[J].电子学报,2009,37(3):622-627.
[23]Canny J. A Computational Approach to Edge Detection[J].IEEE Transactions on PatternAnalysis and Machine Intelligence,1986,PAMI-8(6).
[24]Smith S M, Brady M. SUSAN. A New Approach to Low Level Image Processing[J].International Journal of Computer Vision.1997,23(1).
[25]Lowe D G. Object recognition from local scale-invariant features[C]. International Conferenceon Computer Vision, Corfu, Greece,1999.
[26]Jie Shao et al. Feedback strategy on real-time multiple target tracking in cognitive visionsystem[J]. Optical Engineering50(10),107201(October2011)
[27] Kass M, Witkin A, Terzopoulos D. Snakes: Active contour models[J]. International Journal ofComputer Vision,1988,1(4):321-331.
[28] Malladi R, Sethian J A, Vemuri B C.Shape modeling with front propagation: A level setapproach[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence,1995,17(2):158-179.
[29] Aggarwal J K, Cai Q. Human motion analysis: A review[J]. Computer Vision and ImageUnderstanding,1999,73(3):428-440.
[30] Hu W M, Tan T N, Wang L, Maybank S. A survey on visual surveillance of object motion andbehaviors[J]. IEEE Transactions on Systems, Man, and Cybernetics-Part C: Applications andReviews,2004,34(3):334-352.
[31]王虎元,雷万云,仲照祥等.多模实时图像处理跟踪算法研究[J].弹箭与制导学报,1996(4)：5-10.
[32]陈朝阳,张桂林.多模成像跟踪器的算法智能调度方法[J].华中理工大学学报,1997(12):4-5.
[33]韩客松,刘芳.多态运动目标的智能跟踪策略[J].系统工程与电子技术,1999,21(11):29-31.
[34]董文娟,王宏义,韩建涛,陈曾平.基于SRIO的实时并行图像处理机[J].计算机工程,2009,35(7),20-22.
[35]苏光大.图像并行处理技术[M].北京:清华大学出版社,2002.
[36]李国辉,李亚安,林关成等.基于多片DSP的相关匹配算法的研究与实现[J].弹箭与制导学报,2010,30(1):58-61.
[37]Lei Tao, Jiang Ping, Zhou Jin, Wu Qin-zhang. Study of high speed parallel image processingmachine[J].Proceedings of SPIE-The International Society for OpticalEngineering,2012,3:223-226.
[38]阿列玛索夫BE.火箭发动机原理[M].张中钦,庄逢辰译.北京:宇航出版社,1991.
[39]谭宇.红外成像识别和跟踪技术及其在武器系统中的应用和前景.光电与控制,1998,69(1):21-28.
[40]朱明,宋建中.一种自动识别运动目标前端的方法[J].光学精密工程,1994,2(5):22-27.
[41]彭先蓉,张启衡.高帧频电视跟踪算法的研究[J].光电工程,1998,25(supp):47-50.
[42]张启衡等.高帧频电视跟踪系统的特点与关键技术研究[J].光电工程,1998,25(6):65-69.
[43]蔡敬菊,彭先蓉.适用于快速扩展目标的跟踪算法[J].光电工程,1999,26(supp):91-94.
[44]王坤等.基于局部轮廓特征的无人机头部检测跟踪算法[J].光学技术,2011,37(2):178:182.
[45] J. N. Sanders-Reed, Multi-Target, Multi-Sensor, Closed Loop Tracking[J]. Proc. SPIE5430,1-19(2004).
[46] J. W. Bukley, R. M. Cramblitt, Comparison of image processing algorithms for trackingilluminated targets[J]. Proc. SPIE3692,234-243(1999).
[47]雷涛,蒋平,周进,吴钦章。基于运动方向及主轴的目标头部跟踪点稳定提取[J].光电工程,2012,39(4):49-54.
[48]雷涛,蒋平,周进,吴钦章.一种扩展目标稳定跟踪点提取方法.中国,发明专利, CN102831619A,2012.12.19.
[49]何斌,马天予,王运坚.朱红莲.Visual C++数字图像处理[M].北京:人民邮电出版社,2001,435-437.
[50]张天序.目标提取与跟踪的稳定性研究[J].华中理工大学学报,1994,22(5):55-60.
[51]徐碧,傅承毓.电视跟踪系统中的实时视频图像分割技术[J].光电工程,1999,26(3):8-12.
[52]闫成新.基于区域的图象分割技术研究[D].武汉,华中科技大学,2004.
[53]John N.Sanders-Reed, Mary Jo Duncan, W.B.Boucher, W.Michael Dimmler, Shawn O’Keefe.Multi-Target Tracking In Clutter[J]. SPIE4724April,2002:1-7.
[54]冈萨雷斯.数字图像处理（第二版）[M].北京:电子工业出版社,2003.
[55]赵菲.基于硬件加速的实时二值图像连通域标记算法[J].电子与信息学报,2011,33(5):1069-1075.
[56]张桂林,陈益新,李强等.基于跑长码的连通区域标记算法[J].华中理工大学学报,1994,22(5):11-14.
[54]D.P.Huttenlocher, G.A.Klanderman, W.J.Rucklidge.Comparing Images usiing the HausdorffDistance[J]. IEEE. Trans. On Pattern Analysis and Machine Intellinence,1993,15:850-863.
[55]Viola P, Wells W.Alignment by Maximization of Mutual Information[J]. International Journal ofComputer Vision,1997,24(2):137-154.
[56]倪军,周进,袁家虎,吴钦章.空间目标序列图像相关跟踪算法[J].光电工程,2006,33(4):001-005.
[57]任仙怡,廖云涛,张桂林等.一种新的相关跟踪算法研究[J].中国图象图形学报,2002,7(6):553-557.
[58]王芳.电视制导目标跟踪算法的研究[D].哈尔滨:哈尔滨工业大学,2007:9-11.
[59]R.Venkateswarlu,M.H.Er,Y.H.Gan,Y.C.Fong.Area-correlation tracker with improvedreliability[J].SPIE Vol.3086,1998,244-253.
[60]李俊山,谭园园,张媛莉.SSDA的改进算法[J].光电与控制,2007,14(2):66:68.
[61]李强,张钱.一种基于图像灰度的快速匹配算法[J].软件学报,2006,17(2):216-222.
[62]王博.频域相关的恒常性特征及其算法研究[J].中国图像图形学报,2006,11(6):798-803.
[63] Hu M K. Pattern Recognition by Moments Invariants[A]. Proc. IRE[C],1961.
[64]张坤华,王敬儒,张启衡.基于扩展目标的不变矩跟踪算法.强激光与粒子束.2002(1):6-10.
[65]张凯,刘占文,钟都都,阎杰.红外导引头边缘模板匹配跟踪算法研究[J].西北工业大学学报,2008,26(3).
[66]胡璟璟.复杂场景下目标跟踪的多模板匹配算法研究[D].长沙:国防科学技术大学,2010.11.
[67]罗军,江和平,沈振康.基于匹配跟踪置信度的自适应对应像素距离图像匹配跟踪算法[J].信号处理,2008,24(1).
[68]周进.多模复合图像跟踪技术[D].北京:中国科学院研究生院,2008.
[69]Pan Jiyan, Hu Bo.Robust Object Tracking against Template Drift[C].Proc. IEEE Int. Conf. onImage Processing(ICIP),3:III-353-III-356,2007.
[70] Pan Jiyan, Hu Bo.Robust,Zhang Jianqiu. Robust and Accurate Object Tracking under VariousTypes of Occlusions[J].IEEE Transactions on Circuits and Systems for VideoTechnology(T-CSVT),18(2):223-236,2008.
[71]潘吉彦.具有自适应遮挡处理与模板漂移抑制的视频目标跟踪算法[D].上海:复旦大学,2008.
[72]任仙怡,周晓,张桂林,张天序.基于仿射变换与模板更新策略的目标跟踪方法[J].华中科技大学学报.2001,29,(12):50-51,63.
[73]吉书鹏,张桂林,丁晓青.地面复杂场景图像相关跟踪算法研究[J].激光与红外,2002,32(6).
[74]Wong, S. Advanced correlation tracking of objects in cluttered imagery[J]. Proceedings of SPIE,vol.5810(2005).
[75]Sung-Il Chien, Si-Hun Sung. Adaptive window method with sizing vectors for reliablecorrelation-based target tracking[J]. Pattern Recognition2000(33):237-249
[76]Si-Hun Sung. Adaptive window algorithm with four-direction sizing factors for robustcorrelation-based tracking[J]. IEEE,1997,208-215
[77]Gelfand S B, Fortmann T E, Bar Y. Adaptive Detection Threshold Optimization for Tracking inClutter. IEEE Trans on Aerospace and Electronic Systems,1996.32(2):514-523.
[78]Preston K. Detection of Weak, Subpixel Targets Using Mesh-Connected Cellular Automata.IEEE Trans on Aerospace and Electronic Systems,1990,26(3):548-559
[79]杨卫平,李吉成,沈振康.图像抖动条件下跟踪数据的处理技术[J].电子学报,2001,29(9),pp.1195-1197．
[80]李智勇,沈振康,杨卫平.动态图像分析[M].长沙:国防工业出版社,1998年, pp.151-152.
[81]熊辉,沈振康,魏急波,李吉成.低信噪比运动红外点目标的检测[J].电子学报,1999,27(12),pp.26-29.
[82]李怀琼,陈钱,高文昆.红外焦平面阵列失效元动态检测与校正算法[J].红外与激光工程,2006,35(2), pp.192-196.
[83]黄曦,张建奇,刘德连.红外图像盲元自适应检测及补偿算法[J].红外与激光工程,2011,40(2), pp.370-376．
[84] J Serra.Morphological Filtering: An Over View[J].Signal Processing,1994,38(1), pp.3-11.
[85]叶斌,彭嘉雄,卢汉清.顺序形态滤波在小目标和点目标检测中的应用研究[J].自动化学报,2002,28(6), pp.990-994.
[86]John T.Barnett, Barton D.Billard, Changta Lee.Nonlinear Morphological Processor forPoint-target Detection Versus an Adaptive Linear Spatial Filter: a Performance Comparision[J].SPIE Vol.1954, pp.12-24.
[87]S.Kin, J.Lee.Scale invariant small target detection by optimiazing signal to clutter ratio inheterogeneous background for infrared search and track[J]. Pattern Recognition,2012,45(1):393-406.
[88]X.Z.Bai, F.G.Zhou. Analysis of new top-hat transformation and the application for infrared dimsmall target detection[J]. Pattern Recognition,2010,43(6):2145-2156.
[89]程塨,郭雷,韩军伟,钱晓亮,.基于形态学带通滤波和尺度空间理论的红外弱小目标检测[J].光学学报,2012，32(10)，pp.1-8.
[90]张秀君,孙晓丽.分段线性变换增强的自适应方法[J].电子科技,2005.3:13-16.
[91]周宏仁,敬忠良,王培德.机动目标跟踪[M].北京:国防工业出版社,1991.
[92]罗鹏飞,张文明.一种多目标跟踪航迹起始新算法及其性能评估[J].国防科技大学学报,1999,21(6):51-54.
[93]徐志弘.基于红外全方位警戒系统的多目标跟踪处理[J].舰船光学,1995,2,19-25.
[94]杨卫平,李吉成,沈振康.红外图像序列目标识别与跟踪中的几个问题[J].红外技术,2001,23(4),8-11.
[95]张忠诚,孟庆华,沈振康.红外目标特征分析[J].激光与红外,1999,29(3),166-169.
[96]王晓峰,黄德双,杜吉祥,张国军.叶片图像特征提取与识别技术的研究[J].计算机工程与应用,2006.3,190-193.
[97]于新文,沈佐锐,高灵旺,李志红.昆虫图像几何形状特征的提取技术研究[J].中国农业大学学报,2003,8(3):47-50.
[98]孙即祥.图像分析[M].北京:科学出版社,2005,123-134.
[99]程相权,郭治,余海,戚国庆,李显峰,殷开宝.一种快速目标的图像跟踪综合算法[J].兵工学报,2001,22(4):496:499.
[100]张坤华,王敬儒,张启衡.复杂背景下扩展目标的分割算法研究[J].红外与毫米波学报,2002,21(3):233-237.
[101]王维雅,雷梅,黄向东,谭久彬,李海英.一种小目标快速识别与跟踪方法[J].光电子激光.2007,18(1).
[102]杨宜禾,周维真.成像跟踪技术导论[M].西安:西安电子科技大学出版社,1991.
[103]Javed Ahmed, M N Jafri, Mubarak Shah, et al.. Real-time edge-enhanced dynamic correlationand predictive open-loop car-following control for robust tracking[J]. Machine Vision andApplications,2008(19):1-25.
[104]唐自力，马彩文。基于航迹和特征的目标识别技术在多目标跟踪中的应用[J],光子学报,2010,39(2).
[105]李秋华,李吉成,沈振康,朱振福.基于模糊综合的红外目标融合识别[J].国防科技大学学报.2003,25(1):51-55.
[106]李秋华,基于红外图像信息融合的目标检测和识别技术研究[D].长沙:国防科学技术大学,2002,1.
[107]刘源,谢维信.多传感器图像模糊融合算法在图像识别中的应用[J].西安:西安电子科技大学学报(自然科学版),2000,27(1).
[108]武斌,姬红兵.基于模糊融合的红外弱小目标快速检测算法[J].光电工程,2007,34(12):6-11.
[109]Kerre Etienne E，黄崇福，阮达.模糊集理论与近似推理[M].武汉:武汉大学出版社,2004.
[110]雷涛,周进,吴钦章. DSP实时图像处理软件优化方法研究[J].计算机工程,2012,38(4):177-180.
[111]王鲲鹏,张小虎,李立春,于起峰.一种基于正负差图像的运动目标检测新方法[J].应用光学,2007,28(5):521-525.
[112]董文娟，王宏义，韩建涛等.基于SRIO的实时并行图像处理机[J].计算机工程，2009(7):20-22;
[113]苏海冰.实时图像处理中嵌入式硬件并行结构的研究[D].北京:中国科学院研究生院,2007.
[114] Texas Instruments. TMS320C6455Technical Reference[Z].(2005-12-13). http://www.ti.com.
[115]雷涛,曹小伟,吴钦章.实时DSP图像处理空间低通滤波模块优化[J].光电工程.2012,39(5).
[116]赵丹培,王延杰,李桂菊.基于汇编语言的DSP源代码优化技术的研究[J].电子器件,2004,9(3):459-463.
[117]李晶,国澄明,国狄非等.基于TMS320C64xDSP的HDTV视频解码软件关键模块的优化[J].信号处理,2004,6(3):304-307.
[118]郭元曦,卞红雨.基于TMS320C6416的高频图像声呐算法的实现及优化[J].应用科技,2005,7(7):62-64.
[119]杜歆，陈华华.基于TMS320C64x的立体匹配实时性问题研究[J].传感技术学报，2005,18(2):381-384.
[120]何迪.实时图像跟踪系统中TMS320C6xDSP的软件优化[J].电子科技,2005(4):28-33.
[121]TMS320C6000Optimizing C Compiler User’s Guide[M]. Texas Instruments Incorporated,2006.
[122]TMS320C6000CPU and Instruction Set Reference Guide[M]. Texas Instruments Incorporated,2006.