现实场景的多传感采样与三维重建方法研究

3-D reconstruction for real scenes has wide application prospects in many fields such as battlefield sensing, aerospace exploration, medical image processing, and movie production. Current acquisition approaches and graphic methods cannot characterize the reality, complexity and variability of real scenes, and therefore to achieve accurate and efficient dynamic 3-D reconstruction with low-cost acquisition devices has become a challenging problem. Grasping the opportunity of rapid development and application of Kinect cameras, this project focuses on the sampling and 3-D reconstruction for real scenes via multi-sensors information fusion: based on sparse representation, explore the theory and method of sampling and high-resolution sensing for complex real scenes via sparsely-distributed multi-sensors; study the depth optimization and super-resolution, explore the sensing and tracking of dense features, and investigate the joint spatio-temporal sparse reconstruction for texture features based on the low-rank characteristics of dynamic feature matrix; establish the model of 3-D shape and trajectory for dynamic scenes, and jointly sense the geometry and motion characteristics through the integration of multi-sensor information. We are to reveal the representation mechanism of 3-D characteristics of dynamic real scenes, set up a multi-sensor information sampling and reconstruction platform, and make breakthroughs in both theory and key technology.

现实场景的三维重建在军事、航天、医疗、影视制作等领域有着广泛的应用前景。现有采集机制与图形学处理方法难以刻画现实场景的"真实性"、"复杂性"和"多变性"，采用低成本采集设备实现精确高效的动态三维重建是国际前沿挑战难题。本项目瞄准Kinect体感相机快速发展与应用的契机，开展现实场景的多传感采样与三维重建研究：以信号的稀疏表示为理论依据，探索采用稀疏分布的多传感形式对复杂现实场景进行采样和高分辨率感知的理论与方法；研究深度超分辨率与优化，探索密集特征的感知与跟踪，揭示场景二维投影特征在空间时间维度的运动规律，结合动态特征矩阵的低秩特性研究纹理特征的空时联合稀疏重建；建立动态场景三维形状轨迹模型，并通过融合多传感信息联合感知三维几何与运动特性。力图揭示现实动态场景三维特性的表示机理，建立多传感信息采样与重建平台，在理论和关键技术研究上取得突破。

现实场景的三维重建在军事、航天、医疗、影视制作等领域有着广泛的应用前景。现有采集机制与图形学处理方法难以刻画现实场景的“真实性”、“复杂性”和“多变性”，采用低成本采集设备实现精确高效的动态三维重建是国际前沿挑战难题。本项目瞄准Kinect体感相机快速发展与应用的契机，开展了现实场景的多传感采样与三维重建研究：以信号的稀疏表示为理论依据，探索了采用稀疏分布的多传感形式对复杂现实场景进行采样和高分辨率感知的理论与方法；研究了深度超分辨率与优化，探索了密集特征的感知与跟踪，揭示了场景二维投影特征在空间时间维度的运动规律，结合动态特征矩阵的低秩特性研究了纹理特征的空时联合稀疏重建；建立了动态场景三维形状轨迹模型，并通过融合多传感信息联合感知了三维几何与运动特性。揭示了现实动态场景三维特性的表示机理，建立了多传感信息采样与重建平台，在理论和关键技术研究上取得了突破。共发表学术论文10篇，其中包括SCI二区6篇，SCI三区检索论文2篇，知名国际学术会议2篇,；申请国家发明专利11项，已授权6项。在采用低成本便携设备对现实动态场景进行高效采集、感知与重建等方面提出了原创方法，形成一套多传感信息融合的复杂场景采样与三维重建核心技术。