项目名称: 视频场景下大位移运动目标的持续性跟踪方法研究
项目编号: No.61503173
项目类型: 青年科学基金项目
立项/批准年度: 2016
项目学科: 其他
项目作者: 张焕龙
作者单位: 郑州轻工业学院
项目金额: 20万元
中文摘要: 本项目以视频场景下目标在跟踪过程中产生的运动大位移为研究对象,针对序列图像中可能出现目标自身运动突变、低帧率和镜头切换等现象而引发的复杂动态特性,以及传统算法在实现跟踪过程中因设定运动平滑性约束条件而导致难以适应大位移运动目标持续性跟踪的问题,在算子匹配和积分模型相关理论发展的激励下,从帧间运动场角度研究拓展的SIFT flow方法在无任何先验信息的情况下对大位移运动实现全局性评估的跟踪预测机制,基于最优化、马尔科夫随机场和图像分析等相关理论,以置信度传播、稀疏表示、子空间建模和粒子滤波等方法为核心,建立视频场景中具有大位移运动及模糊、遮挡和形变等外观变化的兴趣目标的持续性跟踪方法和框架,解决大位移运动预测、外观模型构造、学习策略设计和多模型交互方式等问题,以提高视频场景中目标在运动和外观发生显著变化时跟踪算法的精确度和持续性,为跟踪技术在智能监控和行为分析等领域的应用研究提供支持。
中文关键词: 视频跟踪;稀疏子空间模型;大位移运动;凸优化;在线学习
英文摘要: The proposed project addresses the issues on tracking of an object with the large-displacement motion between frames on the sequential images. Considering that the large-displacement caused by abrupt motion, low-frame-rate video and switching camera may destroy the motion smooth assumption so as that the traditional algorithms often track failure, inspired by the both advantages of descriptor matching and integral model we evaluate the large-displacement motion globally without any prior knowledge based on the extended SIFT flow field. With the help of the theories of optimization, markov random fields and image analysis, we construct a long-term tracking framework that not only evaluates large-displacement motion but also adapts the problems of image blur, partial occlusion and image deformation generated by the complex motion using the methods of belief propagation, sparse representation, linear subspace and particle filter. The framework intends to predict motion uncertainty, construct appearance model, design learning strategy and interact multiple model for keeping long-term tracking and improving tracking accuracy, which can promote and guide the development of visual object tracking method in the fields of intelligent surveillance and behavior analysis et al.
英文关键词: visual tracking;sparse subspace model;large displacement motion;convex optimization;online learning