动态场景的时空复用压缩视频成像研究

Compressed imaging(CI) has become one of the hottest research topics in the field of imaging for it breaks the strict limit of the imaging resolution caused by optoelectronic components. However, since the existing compressed imaging methods sample signals by 2D projection, signals are compressed only in a single spatial or temporal dimension. The sampling efficiency and the imaging resolution are still constrained. Aiming at this problem, this project focuses on the theories and methods of spatiotemporal compressed video imaging and its signal reconstruction. The key issues in spatiotemporal compressed video imaging including 3D projection for scene signals, sparse representation and nonlinear reconstruction for video signals will be studied thoroughly in the project. Highlights include: (1) The analog domain compressed sensing(CS) method based on multiply scattering will be explored to combine with the coded exposure technique to solve the 3D projection problem of the instantaneous optical signals. (2) The sparse representation method based on the 3D analytic model for video signals and its dictionary learning algorithm will be investigated. Then, an efficient spatial and temporal sparse representation model of the video data cube will be constructed. (3) The optimization model with multiple prior constraints will be studied to reduce the freedom of the solution of the optimization problem, so that the reconstructed signal tends to the intrinsic characteristics of the video signal. The spatiotemporal compressed video imaging system to be studied eliminates the sampling redundancies in both the spatial and temporal dimensions. It is expected to break the spatial and temporal imaging resolution bottleneck further and to provide a new technical solution for applications on high definition (HD) and high frame-rate (HFR) video acquisition and presentation for dynamic scenes.

压缩成像打破了光电元器件对成像分辨率的严格限制，是当前成像领域的一个研究热点。然而，现有的二维投影压缩成像方法限于单一空间或时间维度上的压缩采样，采样效率和成像分辨率仍然受到制约。针对这一问题，项目研究三维投影的时空复用压缩视频成像及其信号重建理论与方法。项目围绕时空复用压缩成像中场景信号的三维投影、视频信号的稀疏表示和非线性重建等关键问题展开研究，主要创新包括：1.探索采用多重散射的模拟域压缩感知方法结合编码曝光技术，解决瞬时光学信号的三维投影难题；2.研究三维解析模型的视频信号稀疏表示方法与字典学习算法，建立视频数据立方体时空稀疏表示模型；3.构建多先验融合的时空复用压缩感知视频重建优化模型，以降低优化问题解的自由度，引导重建趋向视频信号固有特征。研究的压缩成像方法在时、空两个维度消除采样冗余，以进一步突破时、空成像分辨率瓶颈，为动态场景高清、高帧率视频采集与呈现应用提供新的技术参考。

压缩成像打破了光电元器件对成像分辨率的严格限制，是当前成像领域的一个研究热点。然而，现有的二维投影压缩成像方法限于单一空间或时间维度上的压缩采样，采样效率和成像分辨率仍然受到制约。针对这一问题，项目研究围绕系统中动态场景光学信号的时空复用压缩感知视频成像、视频数据体的三维稀疏表示和视频序列的非线性重构三个核心内容展开，研究三维投影的时空复用压缩视频成像及其信号重建理论与方法。项目提出了一种时空复用的压缩视频成像方法、一种时、空多约束优化的压缩感知视频重建方法、多个低秩优化算法、多种随机数产生方法和其他多种图像、视频特征提取与处理算法，多个算法与当前最优算法相比具有一定优势。项目团队共发表论文15篇，其中SCI检索14篇；申请发明专利7项，其中已授权2项；培养博士2名，硕士10名，很好地完成了预期设定目标。项目为时空复用压缩视频成像研究提供了新的研究思路、理论参考、实验依据，为动态场景的高清、高帧率视频采集与重建应用提供了新的技术方案，具有重要的科学意义和工程应用价值。