视频监控系统在日常生活中有广泛的应用，运动目标视频监测方法是视频监控系统的关键技术之一。复杂背景下感兴趣目标的检测和对感兴趣目标的长时间跟踪是视频监测方法的难点。基于目标局部特征模型的目标检测算法和基于跟踪-学习-检测框架的目标跟踪算法是视频监测方法的重要发展方向之一。 论文对基于隐式形状模型的目标检测算法和基于TLD框架的目标跟踪算法进行了深入研究。论文介绍了隐式形状模型的生成方法和基于隐式形状模型的目标检测算法。针对原算法对目标旋转的适应能力偏弱的问题，在计算投票坐标时加入角度信息，提出了一种对目标旋转有较强适应能力的目标检测算法。论文对目标跟踪中的经典算法——卢卡斯-卡纳德算法，以及最近提出的TLD跟踪算法的框架及思想进行分析与比较，重点研究了将隐式形状模型引入TLD跟踪算法框架的基于隐式形状模型的目标跟踪算法。该算法由跟踪器、检测器、学习器、综合器四个模块组成，算法初始化时完成跟踪目标隐式形状模型的初始建模，检测器基于该模型在输入视频帧中检测目标，跟踪器基于上一帧目标框中的内容在相邻下一帧中搜索目标，综合器融合跟踪器输出结果和检测器输出结果得到跟踪框位置，学习器根据综合器输出结果对目标模型进行更新。四个模块协同工作，完成对监控视频中运动目标的跟踪。 编程实现了本文算法，进行了目标检测与跟踪实验。目标检测实验结果表明改进后的算法对目标角度变化有较强的适应能力。目标跟踪实验结果表明基于隐式形状模型的目标跟踪算法具有较好的抗遮挡性能。本文提出的运动目标检测、跟踪算法在小区、停车场、城市交通等视频监控系统中有很好的应用前景。

Video surveillance system is widely used in daily life. Moving target video monitoring method is one of the key techniques in video monitoring system. Target detection in complex background and long-term target tracking is a difficult point in video monitoring method. Target detection by model of local features and target tracking algorithm based on tracking-learning-detection framework is one of the important development direction. In this paper, the target detection algorithm based on implicit shape model and the target tracking algorithm based on TLD framework are studied. This paper first introduces the method to generate the implicit shape model and the target detection algorithm based on the implicit shape model. We join the angle information in the calculation of coordinate voting, to enhance the ability to adapt the target rotation. Then we introduce the classical algorithm in target tracking—Lucas-Kanade algorithm, as well as the framework and idea of TLD tracking algorithm. On this basis, the implicit shape model is introduced into the TLD tracking algorithm framework. The target tracking algorithm based on implicit shape model consists of a Tracker, Detector, Learner, Integrator. When the algorithm is initialized, the original implicit shape model of tracking object is initialized. Detector based on this model to detect object in the input video frames. Tracker based on the contents of the previous frame in the object frame search object in the adjacent next frame. Integrator fusion tracker output and detector output to get tracking box position. Learner updates the target model according to the results of Integrator. Programming the algorithm proposed in this paper, and the target detection and tracking algorithm performance is tested. Target detection experiment results show that the improved algorithm has the strong adaptive ability to target angle change. Target tracking experiment results show that the target tracking algorithm based on implicit shape model has good anti blocking performance. The target detection and tracking algorithm of this paper has a good prospect in the neighborhood, parking lot, urban traffic video monitoring system.