基于改进XGBoost和随机森林的VR三维手势识别  

VR 3D Gesture Recognition Based on Improved XGBoost and Random Forest

作　　者：邹海洋[1] 李振华 邓利平[1] ZOU Haiyang;LI Zhenhua;DENG Liping(College of Computer,China West Normal University,Nanchong Sichuan 637009,China;College of Education,China West Normal University,Nanchong Sichuan 637009,China)

机构地区：[1]西华师范大学计算机学院,四川南充637009 [2]西华师范大学教育学院,四川南充637009

出　　处：《西华师范大学学报（自然科学版）》2021年第4期426-431,共6页Journal of China West Normal University(Natural Sciences)

基　　金：四川省教育厅资助项目(15ZB0149)。

摘　　要：为了提高VR环境三维手势识别的性能,增强VR交互体验,采用随机森林算法和XGBoost算法进行三维手势识别。根据手掌骨骼轮廓结构、手指及手腕关节点分布,结合三维空间坐标系及关节点相对于X、Y和Z轴的转动角度提取手势数据样本的特征向量,然后采用随机森林算法对特征向量进行降维处理,并且对影响手势识别准确度的特征进行重要度排序,选取重要度排名高的特征向量重构数据样本,最后采用改进的XGBoost算法对重构数据样本进行手势识别训练。实验证明,相比于原始特征向量,随机森林优化后的特征向量训练后三维手势识别准确度更高,且运行时间未有明显增加;而且与常用三维手势识别算法对比,包括SVM、决策树、K-means、神经网络、卷积神经网络,改进的XGBoost算法的手势识别准确率更高。Random forest algorithm and XGBoost algorithm are employed for 3 D gesture recognition so as to improve its performance in VR environment and enhance VR interactive experience.In combination with the three-dimensional coordinate system and the rotation angle of joint nodes relative to X,Y and Z axis,the eigenvectors of gesture data samples are extracted in accordance with the contour structure of palm bone,the distribution of joint points of fingers and wrists.Then,the eigenvectors are reduced by random forest algorithm,and the features that affect the accuracy of gesture recognition are sorted by importance.The ones with high importance are selected to reconstruct the data samples.Finally,the improved XGBoost algorithm is adopted to train hand gesture recognition.Experimental results show that without significantly increasing the running time,the eigenvectors optimized by random forest are more accurate than the original eigenvectors in gesture recognition after the training;moreover,the improved XGBoost algorithm has better performance in gesture recognition accuracy when compared with the commonly used 3 D gesture recognition algorithms such as SVM,decision-making tree,K-means,neural network,and convolutional neural network.

关 键 词：三维手势识别 VR环境 XGBoost算法 随机森林 手掌关节点 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]