本文已被：浏览 3337次 下载 0次
中文摘要: 针对传统行为识别方法存在的数据存储空间不足、识别效率不高以及扩展性不强等问题，本文在利用空间中人体关节点数据进行人体行为表示的基础上，通过自建行为数据集结合Spark MLlib算法库的随机森林算法对行为识别进行建模。为了提升识别模型的泛化能力，本文利用Spark平台下算法的并行且快速迭代的特性，提出了一种多重随机森林的加权大数投票算法。实验结果表明，随着基分类器个数的增加，行为分类准确率显著增高，基分类器个数在5个以后行为识别准确率趋于稳定且高达95%以上。在MSR Daily 3D与MSRC-12数据集上也验证本文行为识别方法的有效性。
Abstract:According to the problems of traditional behavior recognition methods,such as insufficient data storage space, low recognition efficiency and poor scalability, this paper studies how to use the data of human body joint points to represent human behavior, and on this basis, the behavior recognition is modeled by the random forest algorithm of the self built behavior data set combined with the Spark MLlib algorithm library. In order to improve the generalization ability of the model, this paper presents a weighted large number voting algorithm for multiple random forests,using the parallel and fast iterative characteristics of the algorithm under the Spark platform. The experimental results show that with the increase of the number of base classifier, the accuracy of behavior classification is significantly higher, and the recognition accuracy of the base classifier is more stable and up to 95% after 5.The effectiveness of the method also be verified on the MSR Daily 3D and MSRC-12 dataset.
文章编号：cg18000349 中图分类号： 文献标志码：
基金项目:2017年国家重点研发计划项目“矿山安全生产物联网关键技术与装备研发”课题一“矿山‘人、机、环’信息感知和增强现实理论与方法”
引用文本：