基于改进的随机森林的人体部件识别

Body Part Recognization Base on Improved Random Forest

姿态估计是自然人机交互最为重要的环节,人体部件识别是姿态估计的重要步骤。本文介绍了一种基于特征预筛选的改进的随机森林的方法来识别人体各个部件。与传统的随机森林构造不同,在该方法中,对于特征空间十分庞大的实例给出了特征预筛选方法,使得每个分裂节点的特征子集更为高效。该方法既保证了树与树之间的独立,又保证了每棵树的分类性能。在树与树之间的组合中,根据人体部件构造,引入了和分层树的组合模型方式,提高了差异较小类的分类性能,进而提高了森林的准确性。

Pose estimation is the most important step of nature interactive between human and machine, and body part recognition is the core of pose estimation. This paper describes an improved random forests method to recognize each part of the human body. What is different from the traditional random forest structure is that the algorithm proposed in this paper provides a feature Pre - selection for examples with large feature space,making the feature set of each split node more efficient. This method not only ensures the independence between trees,but also ensures classification performance of each tree. In the combina-tion among trees,according to the human part structure,we adopt the combined model of hierarchy forest to improve the classification performance of the forest.