基于运动目标三维轨迹重建的视频序列同步算法

发布时间：2018-03-20 22:23

本文选题：视频同步　切入点：独立运动相机　出处：《自动化学报》2017年10期 　论文类型：期刊论文

【摘要】：提出一种利用运动目标三维轨迹重建的视频时域同步算法.待同步的视频序列由不同相机在同一场景中同时拍摄得到,对场景及相机运动不做限制性约束.假设每帧图像的相机投影矩阵已知,首先基于离散余弦变换基重建运动目标的三维轨迹.然后提出一种基于轨迹基系数矩阵的秩约束,用于衡量不同序列子段间的空间时间对准程度.最后构建代价矩阵,并利用基于图的方法实现视频间的非线性时域同步.我们不依赖已知的点对应关系,不同视频中的跟踪点甚至可以对应不同的三维点,只要它们之间满足以下假设:观测序列中跟踪点对应的三维点,其空间位置可以用参考序列中所有跟踪点对应的三维点集的子集的线性组合描述,且该线性关系维持不变.与多数现有方法要求特征点跟踪持续整个图像序列不同,本文方法可以利用长短不一的图像点轨迹.本文在仿真数据和真实数据集上验证了提出方法的鲁棒性和性能.
[Abstract]:A video synchronization algorithm based on 3D trajectory reconstruction of moving objects is proposed. The video sequences to be synchronized are captured simultaneously by different cameras in the same scene. On the assumption that the camera projection matrix of each frame image is known, the 3D trajectory of moving object is reconstructed based on discrete cosine transform basis, and then a rank constraint based on locus basis coefficient matrix is proposed. It is used to measure the degree of spatial and temporal alignment between subsegments of different sequences. Finally, a cost matrix is constructed, and a graph-based method is used to realize nonlinear time domain synchronization between videos. We do not rely on the known point correspondence. Tracking points in different videos can even correspond to different 3D points, provided they meet the following assumptions: the 3D points corresponding to the tracking points in the observation sequence, Its spatial position can be described by a linear combination of a subset of the set of 3D points corresponding to all tracking points in the reference sequence, and the linear relationship remains unchanged, which is different from most existing methods which require the feature point tracking to persist throughout the image sequence. In this paper, the proposed method can be used to make use of image point trajectories of different lengths. The robustness and performance of the proposed method are verified on the simulation data and real data sets.
【作者单位】：西北工业大学计算机学院;宾夕法尼亚大学工程与应用科学学院;
【基金】：国家自然科学基金(61531014)资助~~
【分类号】：TP391.41

【相似文献】