基于BSCPs-RF的人体关节点行为识别与预测

发布时间：2018-05-09 03:12

本文选题：人体关节点信息 + 行为识别　；参考：《机器人》2017年06期

【摘要】：针对人体关节点序列的连续行为识别问题,提出了一种基于BSCPs-RF(B-spline control points-random forest)的人体关节点信息行为识别与预测方法.首先采用局部线性回归与单帧关节点归一化法预处理关节点序列,以此消除抖动噪声、位移和尺度的影响;然后以B样条曲线控制点作为速度无关的关节点序列特征,并采用同步语音提示词法标注实时行为序列以提高样本采集效率;最后采用基于随机森林的行为识别与预测方法,并以集成学习方法优化多分类器组合以提高识别精度.实验分析了不同参数值对识别效果的影响,并分别在测试数据库MSR-Action3D以及RGB-D设备采集的实时数据集中进行测试.结果显示,MSR-Action3D测试结果优于部分先前方法,而实时数据测试中该方法具有很高的识别精度,进而验证了该方法的有效性.
[Abstract]:In order to solve the problem of continuous behavior recognition of human node sequence, a method of identifying and predicting the information behavior of human node based on BSCPs-RF(B-spline control points-random forestis proposed in this paper. Firstly, the local linear regression and single frame node normalization method are used to preprocess the node sequence, so as to eliminate the effects of jitter noise, displacement and scale, and then the B-spline curve control point is used as the speed independent feature of the node sequence. In order to improve the efficiency of sample collection, synchronous speech prompt lexical tagging is used to improve the efficiency of sample collection. Finally, the method of behavior recognition and prediction based on stochastic forest is adopted, and the combination of multiple classifiers is optimized by integrated learning method to improve the recognition accuracy. The effect of different parameter values on the recognition effect is analyzed experimentally and tested in the real time data set collected by MSR-Action3D and RGB-D equipment. The results show that the MSR-Action3D test results are better than some of the previous methods, and the method has a high recognition accuracy in real-time data testing, which verifies the effectiveness of the method.
【作者单位】：山东科技大学电气信息系;山东科技大学电子通信与物理学院;
【基金】：国家863计划(2015AA016404-4)
【分类号】：TP181;TP391.41

【相似文献】