摘要
智能视频监控是模式识别和计算机视觉领域的主要内容之一,在军事、医学和科研等领域得到了广泛的应用。智能视频监控系统中运动目标检测与跟踪算法的设计是系统的核心,因此,研究智能视频监控系统中的关键技术并提高智能视频监控系统的性能具有重要的意义。
本文针对目标检测与跟踪算法中所存在的主要问题进行了研究,并对相应的问题给出了解决方案。本文的主要研究工作如下:
在目标检测方面,首先对三种常用方法:光流法、背景差分法、帧间差分法进行了分析讨论,并重点研究了三帧差分算法。通过实验表明标准三帧差分算法检测出的运动目标区域不够完整,并且在目标重叠部分不容易检测出来。为了完整、准确地检测出运动目标,本文在原有三帧差分原理基础上进行改进,给出了基于五帧差分的目标检测方法。并通过实验比较了标准三帧差分、增加图像预处理的三帧差分以及五帧差分的检测效果。实验表明,五帧差分法能有效地消除了拖影和空洞现象,更能完整地提取目标,为后面目标特征提取工作打好了基础。
在目标跟踪方面,利用投影法和一阶矩精确地确定了目标的质心位置。给出综合使用目标质心、面积以及矩特征作为目标特征,有效地对目标进行了描述,并给出了基于欧式距离的目标匹配算法。本文利用Kalman滤波器建立运动模型,预测被跟踪目标在下一帧中可能出现的位置,并确定搜索范围,再结合本文给出的目标匹配算法,有效地对运动目标进行了跟踪。最后对多目标中的遮挡问题进行了研究,给出一种在遮挡情况下的目标跟踪方法。对跟踪算法进行了实验仿真,给出了实验数据并对其进行了分析。实验结果表明,基于卡尔曼滤波的目标跟踪模型比较符合实际的情况,跟踪算法能可靠地预测与跟踪目标的运动轨迹。
最后,结合本文改进的运动目标检测方法和基于Kalman滤波器的跟踪模型,采用VC++开发平台和OpenCV算法库设计了一个智能监控的演示系统。通过该演示系统初步验证了本文算法的可行性和可实现性。
Intelligent video surveillance is one of the main content in the field of pattern recognition and computer vision, which has been widely applied in military, medicine and scientific research etc. The design of Moving target detection and tracking algorithm become a core in the intelligent video surveillance system. Therefore, the study of the key technology in intelligent video surveillance system and to improve the performance of the system is of great significance.
The thesis studied the main problem of the target detection and tracking algorithm, and gives the solution to the corresponding problem. The main research work in this article as follows:
On the research of the motion detection, firstly three main algorithms of motion detection which is Optical flow method, the background difference method, inter-frame difference method are researched, the thesis focus on the three differential algorithm. The experimental results show that the region of moving targets aren't integrity, there is a certain shadow, inanition and so on. In order to completely and accurately detect moving targets, the thesis gives a new target detection method based on the five differential, this target detection method is the improvement based on original three differential. The experiment compared the detection effect of the standards three differential, increasing image pre-processing on three differential and five differential detection. The experimental results show the five differential can effectively eliminate the phenomenon of shadow and inanition, more completely extract the target, and make a good foundation for the back of Target Feature Extraction.
On the research of object tracking, the use of projection and the first moment to pinpoint the target's centroid. The thesis integrated use of the target centroid, and moment characteristics of the area as the target characteristics, effectively described the target, and gives the matching algorithm based on the European-style distance. The thesis establish the motion model using Kalman filter, forecast the Possible location in the next frame of the tracked goal, and determine the scope of the search, and then combined with the goal matching algorithm of this paper, effectively track moving targets. Finally, research on the occlusion of more targets, and gives target tracking method in case of shelter. The thesis did the experiment simulation to tracking algorithm, and gave the experimental data and analyzed. The experimental results show that target tracking model accord with actual situation based on Kalman filter, the tracking algorithm can reliably predict the movement and track the target trajectory.
Finally, combined with the improved methods of moving target detection and tracking model based on Kalman filter, using VC + + development platform and OpenCV algorithms library designed a demonstration of intelligent monitoring system. Through the demonstration of the system initially verify the feasibility and achievability of the algorithm.
引文
[1]R.Collins,A.Lipton.Introduction to the Special Section on Video Surveillance.IEEE Transactions on Pattern Analysis.and Machine Intelligence,2000,22(8):745-766
[2]R Collins,A Lipton et al.A system for video surveillance and monitoring:VASAM final report[D].CMU-RI-TR-00-12,Robotic Institute Carnegie Mellon University,2000
[3]Haritao glu I,Harwood D and Davis L.W4:real-time surveillance of people and their activities.IEEE Trans Pattern Analysis and Machine Intelligence,2000,22(8):809-830.
[4]Remagnino P,Tan T,Baker K.Multi agent Visual Survei121ance of Dynamic Scenes[J].Image and Vision Computing,1998,16(8):529- 532.
[5]张文杰,戚飞虎,江卓军,实时视频监控系统中运动日标检测和跟踪的一种实用方法,上海交通大学学报,2002,36(12):1837-1840
[6]刘晓东,苏光大,周全等,一种可视化智能户外监控系统,中国图像图形学报,2000,5(12):1024-1029
[7]周小四,杨杰,朱一坦,用于监控智能报警系统的图像识别技术,上海交通大学学报,2002,36(4):498-501
[8]张艳,王会进,小区智能监控防盗系统的设计与实现,计算机工程与应用,2002,38(11):218-247
[9]谢树煌,陈倩,朱虹,实时视频对象识别与计数系统的模型和算法设计,清华大学学报,2001,41(7):61-64
[10]http://www.goldeneye.net.cn.
[11]http://www.sinocome.com/index.htm
[12]F.Bremond,M.Thonnat.Tracking Multiple Non-rigid Objects in Video Sequences.IEEE Transactions on Circuits and Systems for Video Technology,1998,8(5):585-591
[13]Peihua Li.A Clustering-based Color Model and Integral Images for Fast Object Tracking.Signal Processing:Image Communication,2006,21(8):676-687
[14]G L.Foresti.Object Recognition and Systems and Tracking for Remote Surveillance.IEEE Transactions on Circuits for Video Technology,1999,9(7):1045-1062
[15]Benjamin Coifman,David Beymer,Philip McLauchlan.A Real-time Computer Vision System for Vehicle Tracking and Traffic Surveillance.Transportation Research Part C:Emerging Technologies,1998,6(4):271-288
[16]Stavros Paschalakis,Miroslaw Bober.Real Time Face Detection and Tracking for Mobile Video Conferencing.Real Time Imaging,2004,10(2):81-94
[17]P.Tissainayagam,D.Suter.Object Tracking in Image Sequences Using Point Features.Pattern Recognition,2005,38(1):105-113
[18]Tan T N,Sullivan G D,Baker K D.Model-based Localizations and Recognition of Road Vehicles.International Journal of Computer Vision.1998.222-5.
[19]Meyer F,Bouthemy P.Region-based tracking using affine motion models in long image sequences.CVGIP:Image understanding,1994,60(2):119-140
[20]Bascle B,Deriche R.Region tracking through image sequences.In:Proceedings of IEEE International Conference of Computer Vision,1995,302-307
[21]Paragios N and Deriche R.Geodesic active contours and level sets for the detection and tracking of moving objects.IEEE Trans Pattern Analysis and Machine Intelligence,2000,22(3):266-280.
[22]Isard M and Blake A.Contour tracking by stochastic propagation of conditional density.In:Proc European Conference on Computer Vision,Cambridge,1996,343-356.
[23]Kass M,Witkin A,Terzopoulos D.Snakes:Active contour model s.Internatinal Journal of Computer Vision,1988,1(4):321-331
[24]Polana R and Nelson R.Low level recognition of human motion.In:Proc IEEE Workshop on Motion of Non-Rigid and Articulated Objects,Austin,TX,1994,77-82.
[25]夏良正,李久贤,数字图象处理(第2版),南京:东南大学出版社,2005
[26]章毓晋,图象处理和分析,北京:清华大学出版社,1999
[27]R.Hummel.Image enhancement by histogram transformation[J].Computer Graphics and Image Processing,vol.6,1977:184-195
[28]贾云得,机器视觉[M],北京科学出版社,2000
[29]崔屹编,图像处理与分析数学形态学方法及应用[M].科学出版社,2000.
[30]Franke.U.and S.Heirtrich.Fast Obstacle Detection for Urban Traffic Situations.IEEE Trans.Intelligent Transportation Systems,2002.3(3):173-181P.
[31]Braga-Neto,Ulisses,Goutsias,John.Object-based image analysis using multiscale connectivity[J]IEEE Transactions on Pattern Analysis and Machine Intelligence,v 27,n 6,.June,2005,p 892--907
[32]Herrero E,Orrite C,Senar J.Detected motion classification with a double background and a neighborhood-based difference[J].Pattern Recognition Letters,2003,24(12):2079-2092
[33]Cucchiara R,Piccardi M,Prati A.Detecting moving obiects,ghosts,and shadows in video streams[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2003,25(10):337-1342
[34]Vidal,Rene,Ravichandran,Avinash.Optical flow estimation and segmentation of multiple moving dynamic textures.Proceedings2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition,CVPR 2005,2005,p 516-521
[35]MilIan Sonka,Vaclav Hlavac,Roger Boyle(艾海舟,武勃等译).图象处理分析与机器视觉.第二版.北京:人民邮电出版社,2003.4694910
[36]Robert T.Collins,Alan J.Lipton.A System for'video surveillance and monitoring(A].Conference on Automated Deduction[C],2000:497-501.
[37]李玉山编著,数字视觉视频技术,西安电子科技人学出版社,2006
[38]Rafael C.Gonzalez,Richard E.Woods著,阮秋琦,阮宇智等译,数字图象处理(第二版),北京:电子工业出版社,2003
[39]王坤明,自动选取阈值方法比较研究,抚顺石油学院学报,2002,2.
[40]沈海浪,平西建,许志勇,一种基于背景重建的运动目标检测算法,信息工程大学学报,X01.5(3),Sep.2004
[41]J Bernsen,Dynamic thresholding of grey-level images.Proc.of the 8th Int'l Conf.on Pattern Recognition(ICPR,1986)
[42]Otsu N."A threshold selection method from gray21evel histograms[J]".IEEE Transactions on System Man and Cybernetic,1979,9(1):62-66.
[43]王耀明,陈以,图像阈值分割的统计迭代算法,桂林电子工业学院学报,2000.2.20
[46]Robert T CollinsFFet al.A system for video Surveillance and Monitoring R.Technical Report CMU-RI-TR-00-12,Carnegie Mellon University 2006.
[47]范莹,郭成安,一种运动图像的检测与识别技术,大连理工大学学报,1000-8608-(2004)01-0122-05.
[48]何斌,马天予,王运坚,朱红莲编著,Visual c++数字图像处理,人民邮电出版社,2004
[49]Hu MK.Visual Pattern Recognition by Moment Invariants[J].IRE Trans.on Information Theory,1962.
[50]陈晓荣,陈淑芬,变维卡尔曼滤波实现运动目标的跟踪,仪器仪表学报,2006,27(9),1163-1165
[51]高大山,智能交通系统中运动目标检测方法的研究,[学位论文],清华大学,2002
[52]Yu,Kent K.C,Watson,N.R,Arrillaga,J.An adaptive Kalman filter for dynamic harmonic state estimation and harmonic injection tracking[J].IEEE Transactions on Power Delivery,v 20,n 2 Ⅱ,April,2005,p 1577-1584
[53]钱凤臣,基于图像的目标跟踪中特征提取方法的研究[D],西北工业大学硕士论文,2004.3
[54]叶斌,徐毓,强跟踪滤波器与卡尔曼滤波器对目标跟踪的比较[J],空军雷达学院学报,2005.16(2):17-2
[55]崔峰,冯雪林,彭思龙,近似欧氏距离变换的一种并行算法[z],中国图像图形学报,2004.6
[56]江静,孙晶华,贺利洁,目标识别中的遗传算法研究.佳木斯大学学报,2002,20(1):46-48页
[57]胡硕,基于小波理论的目标检测与快速目标跟踪算法研究,中国科学研究生院博士学位论文,2006
[58]汪颖进,目标跟踪过程中的遮档问题研究[D],华中科技大学硕士论文,2004
[59]张根耀,李竹林,赵宗涛,遮挡情况下运动日标的跟踪[J],安徽大学学报,2003,3(27):30-40
[60]www.Intel.com/technology/computing/opencv/index.htm
[61]刘瑞祯,于仕琪,OpenCV教程:基础篇,北京航空航天大学,2007
[62]陈胜勇,刘盛,基于OpenCV的计算机视觉技术实现,科学出版社,2008.5