单张树木图片中被遮挡枝干的贝叶斯推理及重建

Bayesian Inference and Reconstruction of Sheltered Branches Based on Single Image

采用单张树木图像作为输入,仅需用户简单勾勒,即可自动生成图片中该树木三维可见枝干。为了构建被叶片遮挡的枝干不可见部分,依据树木植物学特征建立了枝条模型样本库,设计了针对不可见枝干建模的评价策略,提出了基于贝叶斯推理的枝干参数优化算法。实验结果表明,基于贝叶斯推理的轻量级树木建模方法能够有效的改进树木三维模型的效果,可为基于单张图片的树木三维建模及优化提供了新的思路。

With a single tree picture as input data, the method could automatically generate the visible 3D branches of the tree. And what the user needed to do was just drawing some strokes ahead along the main branches on the input picture. For the invisible branches, a sample branch model base was reconstructed according to the trees＂ botany characteristics. Some strategies were put forward to evaluate the reconstruction and for a better effect, an optimized algorithm based on Bayesian inference was proposed to improve the reconstruction parameters. The final experiment shows that the method makes a great improvement for 3D modeling of trees, and raises the new approach in this trend.