基于多特征融合的粒子滤波跟踪算法研究

发布时间：2018-02-08 22:42

本文关键词： 多特征融合目标跟踪粒子滤波光照不变图　出处：《西安电子科技大学》2014年硕士论文　论文类型：学位论文

【摘要】：运动目标跟踪是计算机视觉领域的核心研究内容,涉及人工智能、模式识别的前沿研究成果,在军事武器、智能监控、人机交互、虚拟现实等相关领域发挥着无可替代的作用。本文在学习经典的粒子滤波跟踪方法和基于均值漂移的跟踪方法的基础上,对基于多特征融合的粒子滤波跟踪算法进行了比较深入的研究。通过大量实验,发现现有算法在跟踪准确度和精度方面仍存在问题,对有光照变化的场景,跟踪准确度下降,甚至导致跟踪失败,本文从多个方面对算法进行改进,提出了基于加权颜色直方图和边缘直方图的粒子滤波跟踪算法和基于光照不变图的多特征融合粒子滤波跟踪算法。根据人眼视觉敏感性的要求,将目标的空间信息引入到目标特征中,采用加权颜色直方图与边缘直方图相结合的方式表示目标,解决了使用一种特征会导致跟踪失败的问题;对已发散的粒子进行均值漂移迭代,在优化搜索方向的同时缩小搜索范围;此外,对目标跟踪窗口采用自适应调整策略,以提高跟踪过程中目标特征的匹配精度;对目标特征模板进行实时更新,以增强跟踪过程中对抗环境变化的鲁棒性,提升跟踪的准确度和精度。针对光照变化的影响,本文引入了位置敏感直方图,在此基础上,提取视频帧的光照不变图,由于光照不变图对光照变化不敏感,因此将光照不变图应用于基于加权颜色直方图和边缘直方图的粒子滤波跟踪算法中,从而实现目标跟踪,使算法不仅对光照变化具有鲁棒性,而且跟踪准确度和精度也能保持。实验表明本文提出的算法使多特征融合粒子滤波算法的跟踪准确度和精度得到显著提升,而且解决了目标跟踪过程中光照变化带来的跟踪失败问题。与常见目标跟踪算法相比,在中心位置偏移误差和跟踪成功率两个指标上有明显优势。
[Abstract]:Moving target tracking is the core research content in the field of computer vision, involving the advanced research results of artificial intelligence, pattern recognition, in military weapons, intelligent monitoring, human-computer interaction, Virtual reality and other related fields play an irreplaceable role. The particle filter tracking algorithm based on multi-feature fusion is deeply studied. Through a large number of experiments, it is found that there are still some problems in tracking accuracy and precision of the existing algorithms. Even lead to tracking failure, this paper from many aspects to improve the algorithm, Particle filter tracking algorithm based on weighted color histogram and edge histogram and multi-feature fusion particle filter tracking algorithm based on illumination invariant are proposed. The spatial information of the target is introduced into the feature of the target, and the weighted color histogram and the edge histogram are used to represent the target, which solves the problem that the use of a feature will lead to the tracking failure. In order to optimize the search direction and narrow the search range, the target tracking window is adjusted adaptively to improve the matching accuracy of the target features in the tracking process by means of the mean shift iteration of the divergent particles. The target feature template is updated in real time in order to enhance the robustness of the tracking process against environmental changes and improve the accuracy and accuracy of tracking. In view of the influence of light change, the position sensitive histogram is introduced in this paper. Illumination invariant image of video frame is extracted. Because illumination invariant graph is not sensitive to illumination change, Illumination invariant graph is applied to particle filter tracking algorithm based on weighted color histogram and edge histogram to realize target tracking. The algorithm is not only robust to illumination variation, but also can keep the tracking accuracy and precision. Experiments show that the proposed algorithm can improve the tracking accuracy and precision of the multi-feature fusion particle filter algorithm. It also solves the problem of tracking failure caused by the change of illumination in the process of target tracking. Compared with the common target tracking algorithm, it has obvious advantages in center position offset error and tracking success rate.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41

【相似文献】