基于随机森林的三维人造模型前朝向识别算法

摘要

三维模型前朝向的识别是场景合成与重建的基础.针对现有三维模型前朝向识别算法依赖于模型在场景中上下文关系、计算过程复杂等问题,提出一种基于随机森林的三维人造模型前朝向识别算法.首先通过模型简化和计算模型方向包围盒得到候选面;然后基于物体功能设计和心理学等对模型进行形状分析,为候选面提取一组与之相关联的特征;最后利用随机森林理论训练前朝向判别分类器,实现对模型的前朝向识别.实验结果表明,该算法对模型前朝向识别正确率达到80％,能够很好地处理室内场景绝大种类模型,包括目前方法计算错误的模型.%The front orientation detection for 3D models is a fundamental task in interior scenes synthesis and reconstruction.The existing method for 3D models front orientation detection algorithm is content-based and the calculation process is complex.To solve this problem,we propose a front orientation detection algorithm for 3D man-made models based on random forest.We first select candidate bases by model simplification and oriented bounding box computation.Next,we present a set of attributes,defined with respect to each candidate base.The attributes are derived using a combination of functional and psychological considerations.Finally,random forest classifier is trained to determine the front orientation.We have tested the proposed method on different kinds of models in interior scenes.The experimental results show that random forest classifier can achieve sound performance with 80％ accuracy and outperforms the state-of-the-art methods.