闭路电视(closed circuit television, CCTV)系统是内河海事监管的重要手段. 基于跟踪-学习-检测(tracking-learning-detection, TLD)框架研究并改进内河航道CCTV系统的船舶识别和跟踪. 在TLD框架下提出特征值约束条件，可对像素的短期跟踪结果进行校验，不仅有效解决了像素对归一化相关系数值求解的繁琐问题，还很好地保留了图像中角点像素的跟踪结果，使船舶的短期跟踪足够可靠. 用级联的目标检测器精确定位船舶时，在满足内河应用实时性前提下，提出通过对目标候选区域的模板匹配来保证算法准确性. 实验结果表明，改进的算法在应用于内河CCTV系统的船舶识别与跟踪中保持了较高的实时性和鲁棒性，并提高了跟踪精度.
Closed circuit television (CCTV) systems are widely used in the video surveillance of inland waterway. A framework of tracking-learning-detection (TLD) is presented and further enhanced to address the problem of ship identification and tracking in the inland waterway CCTV system. A constraint condition for the
eigenvalues is proposed to examine the short-term tracking results so that complexity in calculating normalized cross correlation is avoided. Meanwhile, tracking results for corners in the image are accurately reserved, making the short-term tracking results reliable. Ships can be located accurately by applying cascaded detectors. A template matching method is proposed to ensure accuracy of the algorithm. Extensive experimental results show that the proposed algorithm outperforms the original TLD framework in terms of efficiency, accuracy and robustness.
[1] 张敬东，王玮. 浅析智能视频技术在海事管理中的应用 [J]. 中国水运，2010, 10(12): 140-154.
ZHANG Jindong, WANG Wei. Analyze intelligent video technology in the application of maritime management [J]. China Water Transport, 2010, 10(12): 140-154.
[2] 郭宝云. CCTV系统目标定位与跟踪技术改进方法的研究 [D]. 大连：大连海事大学，2012.
GUO Baoyun. The improvement about research on object location and tracking technology based on CCTV system [D]. Dalian: Dalian Maritime University, 2012.
[3] 张明杰，李翠华，刘明业. 基于主动轮廓线模型的海面运动目标跟踪 [J]. 计算机工程，2006, 32(6): 34-36.
ZHANG Mingjie, LI Cuihua, LIU Mingye. Ships tracking based on active contour model [J]. Computer Engineering, 2006, 32(6): 34-36. (in Chinese)
[4] 卓炳荣. 海面背景下的红外舰船目标识别跟踪技术研究 [D]. 国防科学技术大学，2011.
ZHUO Bingrong. The research on infrared ship recognition [D]. National university of defense technology, 2011. (in Chinese)
[5] GODEC M, ROTH P M, BISCHOF H. Hough-based tracking of non-rigid objects [C]// IEEE International Conference on Computer Vision, 2011: 81-88.
[6] KALAL Z, MIKOLAJCZYK K, MATAS J. Face-TLD: tracking-learning-detection applied to faces [C]//IEEE International Conference on Image Processing, 2010: 3789-3792.
[7] HORN B K P，SCHUNCK B G. Determining optical flow [J]. Artificial Intelligence, 1981, 17(3): 185-203.
[8] BARRON J L, FLEET D J, BEAUCHEMIN S S. Performance of optical flow techniques [J]. International Journal of Computer Vision, 1994, 12(4): 43-77.
[9] KALAL Z, MIKOLAJCZYK K, MATAS J. Forward-backward error: automatic detection of tracking failures [C]//International Conference on Pattern Recognition, 2010: 2756-2759.
[10] GUO Y L, HSU S, SAWHNEY H S. Robust object matching for persistent tracking with heterogeneous features [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2007, 29(5): 824-839.
[11] KALAL Z, MIKOLAJCZYK K, MATAS J. Online learning of robust object detectors during unstable tracking [C]//IEEE International Conference on Computer Vision, 2009: 1417-1424.
[12] BOSCH A, ZISSERMAN A, MUOZ X. Image Classification using random forests and ferns [C]//IEEE International Conference on Computer Vision, 2007: 1-8.
[13] OZUYSAL M, CALONDER M, LEPETIT V. Fast keypoint recognition using random ferns [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2010, 32(3): 448-461.
[14] LEPETIT V, FUA P. Keypoint recognition using randomized trees [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2006, 28(9): 1465-1479.
[15] KALAL Z, MIKOLAJCZYK K, MATAS J. P-N learning: bootstrapping binary classifiers by structural constraints [C]//IEEE Conference on Computer Vision, 2010: 49-56.
[16] KALAL Z, MIKOLAJCZYK K, MATAS J. Tracking-learning-detection [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2012, 34(7): 1409-1422.
