摘要
目标识别是计算机视觉中一个重要的研究领域。经过几十年的研究和发展,目标识别算法已经取得了显著的理论成果并且广泛应用于社会的各个领域,很大程度地提高了人们的生活质量。随着城市交通设施的不断完善,交通工具数量日益增加,交通压力急剧增大,事故发生率显著上升,智能交通监控势在必行。而智能交通监控是以交通目标识别为前提的,鉴于交通目标识别算法的重要作用和重大价值,对交通目标识别算法的研究和改进具有深远的意义。
针对交通目标的特点,本文展开了对特征提取函数的分析。在此基础之上,进而深入研究了基于SVM的目标识别算法和基于特征点匹配的目标识别算法。由于交通场景的复杂性及交通目标运动的特殊性,分析并比较了多种适合对交通目标进行特征提取的函数。在深入研究支持向量机方法理论和应用原理之后,完成了基于SVM的目标识别算法设计,构造了SVM交通多模式类分类器,通过该SVM多模式类分类器实现对交通目标的识别。实验结果表明该方法有良好的识别效果。此外,实验中还比较了不同特征对目标识别的贡献,这为有效特征的选择提供了依据。在此基础上,提出了一种基于目标识别的卡尔曼多目标跟踪算法,通过目标识别算法解决了卡尔曼多目标跟踪中的质心参数更新的问题,提高了卡尔曼多目标跟踪的准确率。
在基于SVM的目标识别算法的实验中,目标识别的效果会受目标检测结果的影响。为了避免目标检测不准确导致目标识别效果变差的问题,研究了不依赖于目标检测结果的目标局部特征——特征点的提取方法,完成了基于SIFT特征匹配的目标识别算法设计及仿真。SIFT特征以其突出的抗目标几何形变及抗噪声能力在目标识别算法中表现出良好的性能。实验证明该基于SIFT特征匹配的目标识别算法的性能好。
Object recognition is currently one of the most active research topics in computer vision. After decades of research and development, object recognition technique has been developing fast and its wide use is improving the quality of people’s life considerably in various fields. Nowadays, urban transport facilities have been improved and the increasing number of vehicle leads to an increasing accident rate, so intelligent traffic surveillance is imperative. Moreover, traffic object recognition is the groundwork of the intelligent traffic surveillance system. Further research need to be done to improve the object recognition algorithm since it has a significant influence on traffic surveillance.
Considering the characteristics of the traffic object, the feature based object recognition algorithm has been studied. Based on these studies, SVM based object recognition algorithm and object recognition algorithms based on keypoint matching are presented.
Due to the complexity of the traffic scene and the particularity of the moving object, some kinds of feature extraction methods fit for traffic object have been analyzed. After in-depth study of support vector machine theory and its application principles, SVM based object recognition algorithm is proposed, constructing a SVM based multi-modal classifier and then the traffic object can be recognized. Experimental results show that this method is effective. In addition, the comparison of the recognition accurate for different feature extraction methods provides a basis for feature selection. Furthermore, an object recognition algorithm based Kalman multi-object tracking algorithm is presented. The object recognition algorithm resolves the issue of updating the parameters of Kalman filter. Experimental results show that these modifications further improve tracking performance.
However, the results of SVM based object recognition algorithm demonstrate that the effect of object recognition algorithm will be affected by the object detection result. In order to resolve this problem, the local feature extraction algorithm which doesn’t depend on object detection result has been analyzed and the object recognition algorithm based on corner feature matching is proposed. SIFT shows great advantage in dealing with the noises and object distortion. Experimental results indicate that object recognition algorithm based on SIFT feature matching is efficient and accurate.
引文
[1] A. Jain, R. Duin, J. Mao. Statistical pattern recognition: A review. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2000, 22(1): 4-37
[2] M. Hahnel, D. Klunder, K. Kraiss. Color and texture features for person recognition. 2004 IEEE International Joint Conference on Neural Networks, 2004: 647-652
[3] A. Berg, T. Berg, J. Malik. Shape matching and object recognition using low distortion correspondences, IEEE Computer Society Conference on Computer Vision and Pattern Recognition, 2005, 21: 26-33
[4] G. Csurka, C. Dance, L. Fan, et.al. Visual categorization with bags of keypoints. Proceedings of ECCV International Workshop on Statistical Learning in Computer Vision, 2004: 59-74
[5] C. Jian, L. Kan, G. Chunxiao, et.al. Corner-based feature for object recognition. Second International Symposium on Knowledge Acquisition and Modeling, 2009: 106-109
[6] S. Ullman. Object recognition and segmentation by a fragment-based hierarchy. Trends in Cognitive Sciences, 2007, 11(2): 58-64
[7] C. Cortes, V. Vapnik. Support-vector networks. Machine Learning, 1995, 20(3): 273-297
[8] VN. Vapnik, V. Vapnik. Statistical learning theory. Wiley-Interscience, 1998
[9] V. Vapnik. The nature of statistical learning theory. Springer Verlag, 2000
[10] V. Vapnik. An overview of statistical learning theory. IEEE Transactions on Neural Networks, 1999, 10(5): 988-999
[11] J. Zhang, M. Marszalek, S. Lazebnik, et.al. Local features and kernels for classification of texture and object categories: A comprehensive study. International Journal of Computer Vision, 2007, 73(2): 213-238
[12]陈亮. Gabor小波特征提取技术及其在目标识别中的应用研究:[硕士学位论文].南京:南京理工大学,2009
[13]谢松法.模式特征的提取与应用研究:[博士学位论文].武汉:华中科技大学,2007
[14] T. Tuytelaars, K. Mikolajczyk. Local invariant feature detectors: A survey. Foundations and Trends(R) in Computer Graphics and Vision, 2008, 3(3): 177-280
[15] C. Harris, M. Stephens. A combined corner and edge detector. Proceedings of Fourth Alvey Vision Conference, 1988: 147-151
[16] T. Lindeberg. Feature detection with automatic scale selection. International Journal of Computer Vision, 1998, 30(2): 79-116
[17] K. Mikolajczyk, C. Schmid. Scale and affine invariant interest point detectors. International Journal of Computer Vision, 2004, 60(1): 63-86
[18] K. Mikolajczyk, C. Schmid. A performance evaluation of local descriptors. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2005, 27(10): 1615-1630
[19] D. G. Lowe. Object recognition from local scale-invariant features. The Proceedings of the Seventh IEEE International Conference on Computer Vision, 1999, 1152: 1150-1157
[20]程娟.复杂背景下运动目标识别算法研究:[硕士学位论文].武汉:武汉理工大学,2008
[21]陈振学.基于特征显著性的目标识别方法及其应用研究:[博士学位论文].武汉:华中科技大学,2007
[22]盖光建.基于图像的特征信息提取与目标识别:[硕士学位论文].哈尔滨:哈尔滨理工大学,2009
[23] S. Belongie, J. Malik, J. Puzicha. Shape matching and object recognition using shape contexts. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2002, 24(4): 509-522
[24] H. Ming-Kuei. Visual pattern recognition by moment invariants. IRE Transactions on Information Theory, 1962, 8(2): 179-187
[25] J. Flusser, T. Suk. Pattern recognition by affine moment invariants. Pattern Recognition, 1993, 26(1): 167-174
[26]李迎春,陈贺新,高磊.基于仿射不变矩的神经网络目标识别.计算机工程,2004,30(2):31-32
[27] D. Shen, H. Ip. Discriminative wavelet shape descriptors for recognition of 2-d patterns. Pattern Recognition, 1999, 32(2): 151-166
[28]杨蕊红.小波矩算法在图像识别中的应用研究:[硕士学位论文].西安:西北工业大学,2002
[29] R. Gonzalez, R. Woods.数字图像处理(阮秋琦,阮宇智,等,译者).北京:电子工业出版社,2004
[30]王业琴,赵志国.基于直方图和颜色矩方法的木材表面颜色特征的表达.林业科技, 2006,31(5):56-58
[31] M. Partio, B. Cramariuc, M. Gabbouj, et.al. Rock texture retrieval using gray level co-occurrence matrix. Proceedings of 5th Nordic Signal Processing Symposium, 2002: 8-4
[32]刘忠伟,章毓晋.综合利用颜色和纹理特征的图像检索.通信学报,1999,20(5):36-40
[33]杨镇宇.城区复杂背景条件下的目标识别算法研究:[硕士学位论文].重庆:重庆大学,2009
[34]万琴,王耀南.基于卡尔曼滤波器的运动目标检测与跟踪.湖南大学学报:自然科学版, 2007,34(3):36-40
[35]姚红革,耿军雪.基于卡尔曼预测的视频目标实时跟踪.西安工业大学学报,2007,27(2):171-175
[36]史廷彦,赵书斌.基于不变矩和角点特征的目标识别.指挥控制与仿真,2008,2:32-34
[37]刘萍萍.图像的局部不变性特征方法研究:[博士学位论文].长春:吉林大学,2009
[38] L. Kitchen, A. Rosenfeld. Gray-level corner detection. Pattern Recognition Letters, 1982, 1(2): 95-102
[39] H. Wang, M. Brady. Real-time corner detection algorithm for motion estimation. Image and Vision Computing, 1995, 13(9): 695-703
[40] S. Smith, J. Brady. Susan-A new approach to low level image processing. International Journal of Computer Vision, 1997, 23(1): 45-78
[41] D. Lowe. Distinctive image features from scale-invariant keypoints. International Journal of Computer Vision, 2004, 60(2): 91-110
[42]郭佳欣.基于SIFT和NDLT的目标识别技术研究:[硕士学位论文].太原:中北大学,2009