跟踪器融合的视觉跟踪方法研究

项目名称： 跟踪器融合的视觉跟踪方法研究

项目编号： No.61303104

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 丁辉

作者单位： 首都师范大学

项目金额： 23万元

中文摘要： 视觉跟踪是计算机视觉领域的一项重要研究课题，当前的视觉跟踪方法在处理目标运动和目标外观的复杂变化时还存在诸多困难。本课题面向复杂环境下的视觉跟踪问题，从跟踪器融合的角度研究一种新的视觉跟踪方法，包括：1）研究建立一个评价跟踪器集的度量准则，并提出一种构建跟踪器集的在线方法，使得跟踪器集能够自适应地描述跟踪环境的复杂变化；2）研究一种并行随机逼近蒙特卡洛采样算法对复杂环境下的多模滤波分布进行高效地采样，进而构建多条并行交互马氏链实现自适应的跟踪器融合，使得跟踪系统能够持续地适应跟踪环境的复杂变化，从而提高跟踪系统的鲁棒性和准确性；3）设计和实现一个跟踪系统原型，并验证和评估跟踪方法的有效性。

中文关键词： 视觉跟踪器；外观模型；多信息融合；贝叶斯理论；

英文摘要： Visual tracking is an important research task in computer vision community. It is still challenging for existing methods in dealing with the abrupt changes in motion and appearance. In this work, a novel method based on tracker fusion is studied for tracking in complex real-world environment. Firstly, a new measurement criterion is studied and built for evaluation of the trackers set, through which an online method is proposed for constructing the trackers set that can adaptively characterize the real-world tracking environment that varies severely over time. Secondly, a parallel SAMC sampling algorithm is studied to efficiently sample from the multimodal filtering distribution, through which several parallel Markov chains with interactions are built for adaptive trackers fusion, which adapts the tracking system to the tracking environment that varies severely over time, and thus improves the tracking robustness and accuracy. Finally, A prototype tracking system is built based on the proposed method, and the experimental evaluation is performed to validate the efficacy of the proposed method.

英文关键词： Visual tracker；Appearance model；Multiple information fusion；Bayesian theory；