一种动态稀疏式观测系统的场景重建方案

A Scene Reconstruction Scheme for Dynamical Sparse Mode Observation System

为使单目视觉系统获得动态观测并密集重建场景内大型目标的能力,提出一种使用动态观测系统重建场景的方法,设计一套用于实现的系统方案。采用稀疏式的数据更新方法,维护观测数据以降低更新率,从而达到削减系统负担的目的。拟合参考平面,以筛选数据内点进行外点抑制。细分多边形初始化模型,通过混合建模的方法局部优化模型曲面,以达到稀疏到密集重建目标结构模型的目的。实验结果表明,相对于tM=tN=0.65时,当维护判定参数tM=tN=0.5时,数据更新次数由9次降为6次,观测点集由4154降为3414,相应的外点比例由20.94%下降到17.19%。在2次细分的条件下,最终模型片元数由86712减少为71 631,降低了系统负担,并且重建的视觉效果更佳。

For achieving the ability to dynamically observe and densely reconstruct the large-scale object in the scene, a method is proposed to reconstruct the scene by using a dynamical observation system, and a corresponding system solution is designed to implement the method. By using sparse mode data-updating method, the target of reduction of system burden can be achieved by maintaining the observed data which can reduce updating ratio. The outliers can be suppressed by selecting the inliers with the referenced plane. By subdividing the initial polygon model and applying the hybrid modeling method to fit the local surface of the model, the target of objective structure reconstruction can be achieved from sparse to dense. The result shows that, by applying the updating times can be reduced from 9 times to 6 times, the observed points can be reduced from 4 154 to 3 414 and corresponding outliers ratio can be reduced from 20.94% to 17.19%. Under the condition of twice subdivisions, the ultimate number of cells on model can be reduced from 86 712 to 71 631, which can reduce the system burden and achieve the better visual effect on reconstruction.