基于优化DRG的三维人体点云骨架提取方法

Research on method of extracting skeletal curves of 3D human scanning data based on optimizing DRG

针对离散Reeb图(Discrete Reeb Graph,DRG)描述人体骨架时分支部位骨架线偏离中轴的问题,采用了能量函数最小化的方法对DRG曲线进行优化。将人体模型的DRG曲线作为初始骨架,定义其能量函数,在点云模型的距离场梯度的作用下,迭代地调整偏离中轴目标段的曲线位置使其逐渐逼近中轴,能量函数最小时得到优化的骨架。将该算法应用于同一模特四个不同姿势和四个不同模特同一姿势的人体点云模型,并与基于拉普拉斯算子的点云收缩的骨架提取方法进行了比较。结果表明,该算法能够很好地适应各种不同姿势和体型,模型分叉部位的特征得到更加完善的描述,得到的骨架曲线更接近模型的中轴。

By minimizing the energy function, the optimization algorithm on the Discrete Reeb Graph（DRG）is put forward to reduce the deviation of the branch skeletal curves from the medial axis. The proposed method takes the DRG of the model as the initial skeleton curve, and later defines an energy function to deal with deviation. The distance field gradient of the scanning data leads the initial skeleton curve to the axis position step by step through iteratively adjusting positions of target curve, and the optimized skeletal curves can be realized when the energy value of the curves is minimized. In the experiments, the proposed method is applied to the scanning data of the same model under four different postures and four different models with the same posture respectively, and it also compares this algorithm with curve skeleton extraction via Laplacian-based contraction. The results verify that the optimized algorithm proposed can adapt to different postures and different body types, and ensure that ultimate curves are closest to the medial axis. Moreover, better descriptions on fea-tures of furcation curves can also be obtained.