基于多光谱视频的目标跟踪技术研究

项目名称： 基于多光谱视频的目标跟踪技术研究

项目编号： No.61203266

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 自动化学科

项目作者： 赵高鹏

作者单位： 南京理工大学

项目金额： 24万元

中文摘要： 传统视频目标跟踪技术仅利用单一光谱的视频传感器，跟踪精度、稳定性和应用场合受传感器成像特性的限制，难以实现全天时全天候的稳定准确跟踪。本课题基于多光谱视频的目标跟踪技术，深入研究多光谱视频的配准问题、基于多光谱视频的复杂环境下目标跟踪问题。主要研究内容包括：通过研究基于熵和结构图像表示的多光谱视频局部不变共有特征检测描述算子，提出一种多光谱视频配准算法；利用经过配准的多光谱视频进行目标跟踪，通过研究一种基于分层的特征融合跟踪框架，将目标的多光谱特征模型描述、上下文跟踪和目标属性特征关联有机结合，设计视频单/多目标跟踪算法，实现目标稳定准确跟踪；最后通过CUDA编程实现配准和跟踪算法的GPU并行处理。本课题通过利用多光谱视频的冗余和互补信息，为提高视频跟踪系统性能提供一种思路和技术支撑，预期成果可广泛应用于智能视频监控、战场侦察监视等领域，选题具有重要的理论意义和应用价值。

中文关键词： 多光谱视频；目标跟踪；目标检测；压缩感知；并行处理

英文摘要： Traditional video object tracking technology uses only a single spectrum video sensor, so the tracking precision, stability and applications are limited by the sensor imaging characteristics, and it is difficult to achieve stable, accurate tracking for all-time, all-weather conditions. The issue proposes the research on object tracking based on multi-spectral video. The key problems will be studied including multi-spectral video registration and object tracking based on multi-spectral video in complex scenes. The specific research topics include the following aspects. A multi-spectral video registration algorithm is proposed by studying local invariant common feature detector and descriptor based on entroy and structural image representation, and a hierarchical feature fusion tracking framework is proposed by fusing the object multi-spectral video registation, context tracking and object attribute characteristics, the single object tracking algorithm and multi-object tracking algorithm are proposed based on this framework to achieve stable and accurate tracking in complex scenes. Finally the GPU parallel processing of the registration and tracking algorithms is achieved by CUDA programming.The issue can provide theoretical and technological support for improving video tracking system performance through the us

英文关键词： multi-spectral video；object tracking；object detection；compressive sensing；parallel processing