多视图像序列中运动目标分割优化  

Real-time Moving Object Segmentation Based on Multi-View Images Sequence

作　　者：陈国军[1] 陈燕 韦鑫[1] 

机构地区：[1]中国石油大学(华东)计算机与通信工程学院,青岛266580

出　　处：《系统仿真学报》2014年第9期2023-2027,共5页Journal of System Simulation

基　　金：国家"973"计划项目(2009CB320805)

摘　　要：在多视图像的三维重建中,多视图像序列中的运动目标轮廓提取影响重建物体的准确性及重建实时性。针对多视图像间的空间约束关系,提出基于图像分块和并行计算的优化效率算法。算法对图像进行合理分块,确定多视图像的有效成像区域,区分冗余块和非冗余块,基于非冗余块进行优化加速计算,结合连续帧之间的相关性,将帧间连贯性应用于运动目标分割,进一步优化检测算法。将分割过程进行并行处理。实验表明:提出的方法在保证分割准确性的前提下,可以较好地提高分割计算效率,保证了三维重建的实时性。In the three-dimensional reconstruction based on multiview images, moving object contour extraction of multi-view images sequence affects the accuracy and real-time nature of the reconstruction. A parallel optimization method was proposed based on image blocks, considering the space constraint relations of mnltiview images. The image was divided into blocks reasonably. The effective imaging area of multiview images was calculated to distinguish redundant and non-redundant image blocks, thus optimizing calculation based on non-redundant blocks. Combining with the correlation between successive frames, the interframe consistency was used in moving object segmentation to further optimize the detection algorithm. The whole segmentation process was performed in parallel. According to the experiments, the proposed method can better improve the segmentation efficiency, and enhance the real-time nature of 3D reconstruction in the premise of ensuring the segmentation accuracy.

关 键 词：运动目标分割 多视图像 图像分块 帧间连贯性 并行计算 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]