摘要
自动视觉监控是计算机视觉的一个重要研究领域，它在交通信息监测以及银行、宾馆、车站等重要场所的监控中有广泛的应用前景。
本文以智能化视频道路交通监测系统中运动车辆的检测与跟踪为应用背景，对运动目标检测中背景更新、阴影和车灯投影噪声的消除等一些难点问题进行了深入的分析研究，提出了一个稳定的运动车辆检测和跟踪的算法，并在此基础上完成了实用的视频道路交通监测系统的设计与开发。主要内容如下：
1． 基于户外自然光照系统的特点，建立了基于图像差分直方图和递推最小二乘滤波（Recursive Least Square Filter,简称RLS滤波器）的局域化背景和光照变化模型，提出了自适应的背景判断和更新算法。实验结果表明算法可以有效抑制噪声，能够稳定地跟踪实际环境中自然光照条件的变化。
2． 针对阴影干扰的消除，提出稳定的运动车辆检测算法，从图像差分直方图中提取特征，应用具有很好扩展性的支持向量机（Support Vector Machine,简称SVM）分类器进行阴影与车辆的分类，实验结果令人满意。本文还将此方法用于夜间车灯投影干扰的检测与消除中，也取得比较好的效果。检测算法还对检测到的车辆区域采用平行四边形的轮廓描述，能很好代表运动车辆的形状，保证了车辆信息提取的可靠性。
3． 将车辆检测算法应用于运动车辆轨迹跟踪中，我们采用基于车辆均加速运动假设的Kalman滤波器模型。算法具有模型简单实用，检测计算量小，对环境变化适应性强的特点，检测与跟踪方法的相互校正提高了系统的可靠性。
4． 完成了实用的视频道路交通监测系统，并对系统进行实时性、准确性和稳定性的评测。评测结果表明，该系统能够全天候地对交通车流进行实时监测，提供全面准确翔实的统计数据，达到了实用标准，可以为交通信息系统提供可靠的交通统计数据。
本文的运动车辆检测和跟踪方法作为视觉监控领域的一种普遍方法，具有一定的理论意义和实用价值，可以推广到视觉监控的其他应用中，具有广阔的应用前景。
Automatic video surveillance is an important research area in computer vision. It has promising prospect in the applications of collecting traffic information, and monitoring the important areas, such as bank, hotel, and railway station.
Upon the background of detecting and tracking moving vehicles in an intelligent video-based road traffic surveillance system, in this thesis, we focus our research on some hard issues in detecting moving objects, such as updating background image, eliminate the disturbances of moving cast shadow and vehicle headlight, and propose a robust algorithm of detecting and tracking moving vehicles on road. We also apply our algorithm to an applicable road traffic monitoring system. The main contents can be listed as follows:
1. After analyzing the properties of the natural illuminating system, a region-based background and natural illumination model is set up using the histogram of image difference and RLS filter. On the basis of that model, we propose an adaptive algorithm of judging and updating background image. The experiment indicates that the algorithm can suppress noise effectively and adapt to the variation of the illuminating condition robustly.
2. Aiming at eliminating the disturbance of moving cast shadow in detection, a robust algorithm of detecting moving vehicles is proposed. By choosing the histogram of image difference as features, a Support Vector Machine (SVM)-based classifier, which has high generalization performance, is used to segment shadows and vehicles. The result of experiment is rather satisfying. We also apply the classifier to eliminate the disturbance of vehicle headlight at night and it also works. In the moving vehicle detection algorithm, the shape of a vehicle is well represented by a parallelogram, which makes the information of the vehicle extracted from the image reliable.
The detection algorithm is applied in the vehicle tracking. By assuming that the vehicle moves with constant acceleration from current frame to the next, a Kalman filter model is used to tracking and predicting the trace of a vehicle. The detection algorithm has some advantages such as an explicit and applicable
3. model, the low computation cost during detection and its robust adaptation to the variation of environment. The interaction between the phases of detection and tracking improves the reliability of the whole algorithm.
4. An applicable road traffic monitoring system is developed on the basis of our detection algorithm. The design of the system architecture, functional modulates and user interfaces is introduced in the thesis. The system is tested for its real-time performance, reliability and stability and the experiment shows that such traffic monitoring system can work real-timely under any weather and illuminating condition, providing comprehensive and accurate statistics data of traffic flow. It satisfies the demand of real environment application.
The proposed algorithm of detecting and tracking moving vehicles in this thesis, which has important theoretic meaning and application value, is a universal method in the research area of video surveillance. It can be easily generalized to the applications with different situations, and will have a wide application prospect as well.
引文
[1] 高文，陈熙霖.计算机视觉--算法与系统原理. 北京：清华大学出版社，广西科学技术出版社，1999. 83-95
[2] 王栓，艾海舟，何克忠.基于差分图像的多运动目标的检测与跟踪，中国图像图形学报，1996，4(6):470-475
[3] J.Bulas-Cruz, A.T. Ali and E.L. Dagless. A temporal smoothing technique for real-time motion detection. Proceedings of ECCV '94. New York: Springer-Verlag, 1994. 379-386
[4] D. Koller, K. Daniilidis, T. Thorhallson, et al. Model-Based object tracking in traffic scenes. Proceedings of ECCV '94, New York: Springer-Verlag, 1994. 437-452
[5] Soo-Chang Pei and Lin-Gwo Liou. Vehicle-type motion estimation by the fusion of image point and line features. Pattern Recognition, 1998, 31(3): 333-344
[6] Yuncai Liu and Thomas S. Huang. Vehicle-Type motion estimation from multi-frame images. IEEE Transactions on pattern analysis and machine intelligence, 1993, 15(8): 802-808
[7] N. J. Ferrier, S. M. Rowe and A. Blake. Real-Time traffic monitoring. Image and vision computing, 1995, 13(5): 359-367
[8] Simon Rowe, and Andrew Blake. Statistical mosaics for tracking. Image and vision computing, 1996, 14(8): 549-564
[9] K. Karmann, A. Brandt, R. Gerl. Moving object segmentation based on adaptive reference images. Signal Processing: Theories and Applications, 1990, 5(9): 951-954
[10] M. Boninsegna, A. Bozzoli. A Tunable algorithm to update a reference image. Signal Processing: Image Communication, 2000, 16(4): 353-365
[11] A. Lai, N. Yung. A fast and accurate scoreboard algorithm for estimating stationary backgrounds in an image sequence. Proceedings of IEEE International Symposium on Circuits and Systems, 1998, 241-244
[12] R. Cucchiara, C. Grana, M. Piccardi, et al. Statistic and knowledge-based moving object detection in traffic scenes. Proceedings of IEEE Intelligent Transportation Systems, 2000. 27-32
[13] Y.K. Jung, Y. S. Ho. Traffic parameter extraction using video-based vehicle tracking. Proceedings of IEEE/IEEJ/JSAI International Conference on Intelligent Transportation Systems, 1999. 764-769
[14] S. Kamijo, Y. Matsushita, K. Ikeuchi, M.Sakauchi. Traffic monitoring and accident detection at intersections. Proceedings of IEEE/IEEJ/JSAI International Conference on Intelligent Transportation Systems, 1999. 703-708
[15] C. Stauffer, W. Eric L. Grimson. Learning patterns of activity using real-time tracking. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2000, 22(8): 747-757
[16] R.E. Kalman. A new approach to linear filtering and prediction problems. IEEE Trans. on ASME, J. Basic Eng. 82D, 1960, 35-45
[17] Simon Haykin. Adaptive Filter Theory. 3rd edition, Prentice-Hall, Inc. 1996
[18] D. Koller, K. Danilidis, and H.H. Nagel. Model-based object tracking in monocular image sequences of road traffic scenes. International Journal of Computer Vision, 1993, 10(3): 257-281
[19] J. Stauder, R. Mech, J. Ostermann. Detection of moving cast shadows for object segmentation, IEEE Transactions on Multimedia, 1999, 1(1): 65-76
[20] M. Kilger. A shadow handler in a video-based real-time traffic monitoring system. Proceedings of IEEE Workshop on Applications of Computer Vision, 1992. 11-18
[21] Z. Zhu, G. Xu. VISATRAM: a real-time vision system for automatic traffic monitoring. Image and Vision Computing, 2000, 18(10): 781-794
[22] I. Mikic, P.C. Cosman, G.T. Kogut, et al. Moving shadow and object detection in traffic scenes. Proceedings of International Conference on Pattern Recognition, 2000. 321-324
[23] R. Cucchiara, C. Grana, M. Piccardi, et al. Improving shadow suppression in moving object detection with HSV color information. Proceedings of IEEE intelligent transportation Systems Conference, 2001. 334-339
[24] V. Vapnik, Statistical Learning Theory. New York: Wiley, 1998.
[25] C. B. Barber, D.P. Dobkin and H.T. Huhdanpaa. The Quickhull Algorithm for Convex Hulls. ACM Transactions on Mathematical Software, 1996, 22(4): 469-483.
[26] 张有为. 维纳与卡尔曼滤波理论导论. 北京：人民教育出版社, 1980.
[27] S.M.鲍齐克. 数字滤波和卡尔曼滤波. 北京：科学出版社, 1984. 137-150
[28] H.L.Van Tress. Detection, estimation, modulation theory. New York: John Wiley & Sons, 1968. Part I
[29] Zhimin Fan, Jie Zhou and Dashan Gao. Robust Contour Extraction for Moving Vehicle Tracking. To be appeared in IEEE International Conference on Image Processing 2002
Won Kim, Sun-Gi Hong and Ju-Jang Lee. An active contour model using image flow for tracking a moving object. Proceedings of International Conference on Intelligent Robots and Systems, 1999, 216-221