复杂场景下基于状态感知的视频运动目标跟踪研究

Video moving object tracking is an important research topic in the field of computer vision. Existing methods for video moving object tracking are based on low-level feature processing, resulting in low accuracy and robustness in cases of background noises, occlusions and similar objects disturbances. In this project, we propose a moving object tracking method based on state perception in complex scenes. Our method could accurately describe the tracked object by combining local and global features, which constructs new feature-extracting operators reflecting the state change of the object. According to the feature changing characteristics, the object's state could be percept, based on which, effective features are selected qualitatively. To sum up, our method effectively combines low-level features and high-level state knowledge, and improves the robustness of the tracking system in complex scenes. In addition, we propose a similarity measure combining the discrimination between the object and background, which could decide the weights of fused features adaptively. As such, our multi-feature fusion moving object tracking method could deliver more accurate location. In conclusion, our method overcomes the limitation of pixel-level & low-level feature tracking, and provides an accurate and robust video moving object tracking system.

视频运动目标跟踪是计算机视觉领域的重要研究课题。当前方法主要对中低层特征处理，在背景噪声、遮挡、相似物干扰等复杂场景下存在稳健性差、准确性低等问题。本项目提出复杂场景下基于状态感知的运动目标跟踪方法，通过建立全局与局部特征结合的表观模型实现目标的准确描述，构造反映目标状态变化的特征提取新算子，根据特征变化感知目标状态并选择有效特征，实现目标低层特征与状态高层知识的有机结合，提高复杂场景下目标跟踪的稳健性；利用目标与背景区分的相似性度量自适应确定特征的权值，实现多特征融合的准确目标定位。该方法突破了基于像素级低层特征进行目标跟踪的局限，能实现复杂场景下准确、稳健的视频运动目标跟踪。

视频运动目标跟踪是计算机视觉领域的重要研究课题。当前方法主要对中低层特征处理，在背景噪声、遮挡、相似物干扰等复杂场景下存在稳健性差、准确性低等问题。本项目提出复杂场景下基于状态感知的运动目标跟踪方法，通过建立全局与局部特征结合的表观模型实现目标的准确描述，构造反映目标状态变化的特征提取新算子，根据特征变化感知目标状态并选择有效特征，实现目标低层特征与状态高层知识的有机结合，提高复杂场景下目标跟踪的稳健性；利用目标与背景区分的相似性度量自适应确定特征的权值，实现多特征融合的准确目标定位。该方法突破了基于像素级低层特征进行目标跟踪的局限，能实现复杂场景下准确、稳健的视频运动目标跟踪。其研究成果不仅可用于民用领域，也可用于军事领域，实现重要目标的定位打击。