基于视觉显著性的运动目标跟踪方法研究

发布时间：2018-08-09 11:00

【摘要】：运动目标跟踪是在连续变化的图像序列中找出目标的位置和状态的过程,而目标跟踪实现的稳定性和鲁棒性则是要处理的主要问题。由于现实环境的复杂多样性,一般算法对运动目标大小和形状的改变适应性较差,不能有效的解决复杂情况和突变运动下目标跟踪的问题,当目标发生遮挡、目标移动太快以及目标丢失等突变运动时不能自动恢复从而导致跟踪失败,很难实现目标跟踪的准确性和稳定性。为解决突变运动下的目标跟踪问题,本文提出一种基于视觉显著性的运动目标跟踪算法,该算法将视觉注意机制运用到运动目标跟踪框架中,利用时空显著性算法对视频序列进行检测,生成视觉显著图,从视觉显著图对应的显著性区域中建立目标的特征表示模型来实现运动目标的跟踪。论文做了以下工作:(1)对运动目标跟踪算法框架和视觉显著性技术的理论基础进行了阐述,并对目前常用的运动目标跟踪算法有均值漂移法、粒子滤波法、卡尔曼滤波法,和视觉显著性检测算法有Itti、CA、SR、LC进行了分析和实验。(2)通过分析目前主流的时空显著性检测算法,有PQFT和SEG算法,并引入到运动目标跟踪算法框架中,然后进行算法设计来对运动目标进行跟踪。(3)采用国际公共视频序列进行运动目标的跟踪遮挡测试,旨在运动目标发生丢失、遮挡等突变运动情况下和复杂环境下能否准确和稳定的跟踪目标,并与目前主流目标跟踪算法进行实验对比和定量分析。实验结果表明,本文方法在摄像机摇晃等动态场景下可以较准确检测出时空均显著的目标,有效克服了在运动目标发生丢失和遮挡等复杂和突变情况下跟踪不稳定问题,具有较强的鲁棒性,从而实现复杂场景下目标较准确的跟踪。
[Abstract]:Moving target tracking is the process of finding out the position and state of the target in a continuously changing image sequence, and the stability and robustness of target tracking are the main problems to be dealt with. Because of the complexity and diversity of the real environment, the general algorithm has poor adaptability to the change of the size and shape of the moving object, so it can not effectively solve the problem of target tracking under the complex situation and sudden motion. It is difficult to achieve the accuracy and stability of target tracking because it can not recover automatically when the target moves too fast or when the target is lost. In order to solve the problem of moving target tracking under sudden motion, a moving target tracking algorithm based on visual saliency is proposed in this paper, which applies visual attention mechanism to moving target tracking framework. Using spatio-temporal salience algorithm to detect video sequence and generate visual saliency map, the target feature representation model is established from the salience region corresponding to visual salience map to achieve moving target tracking. The following works are done in this paper: (1) the frame of moving target tracking algorithm and the theoretical basis of visual salience technology are expounded. The commonly used moving target tracking algorithms are mean shift method, particle filter method, Kalman filter method, and so on. And visual salience detection algorithms are analyzed and experimented. (2) by analyzing the current mainstream spatio-temporal salience detection algorithms, there are PQFT and SEG algorithms, which are introduced into the framework of moving target tracking algorithm. Then the algorithm is designed to track the moving target. (3) the international common video sequence is used to track the moving object in order to lose the moving target. Whether the target can be tracked accurately and stably under the condition of sudden motion such as occlusion and complex environment, and compared with the current mainstream target tracking algorithm, the experiment and quantitative analysis are carried out. The experimental results show that the proposed method can accurately detect spatio-temporal targets in dynamic scenes such as camera shaking, and can effectively overcome the problem of tracking instability in complex and abrupt situations such as loss and occlusion of moving targets. It has strong robustness so that the target can be tracked accurately in complex scene.
【学位授予单位】：昆明理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】