基于改进ReliefF算法的哑铃动作识别

为了实现哑铃动作分类识别的目标，在哑铃上加装惯性传感器模块，通过采集哑铃锻炼过程中的运动信号，经信号标准化、滤波、基于初始静态量周期分割预处理后，提取侧平举、前平举、反握弯举、锤式弯举、弯举5种哑铃动作的特征向量，使用改进的ReliefF特征选择算法，选择最优特征向量，采用基于平衡决策树的支持向量机对不同的哑铃动作进行分类识别。通过在实验室自主研发的哑铃动作识别系统上进行测试，结果表明：系统能够在单个哑铃动作周期内对哑铃动作进行识别，且识别率可达90%以上，为提供更加个性化的哑铃动作指导奠定基础

Dumbbell motion recognition based on improved ReliefF algorithm

In order to achieve the goal of dumbbell movement classification and recognition,an inertial sensor module is installed on the dumbbell. Through collecting the motion signals during the dumbbell exercise,the eigenvector of five kinds of dumbbell movement,such as flat lifting,counter-grip bending,hammer bending,and curling are extracted after signal standardization,filtering and periodic segmentation based on initial static variables.The improved ReliefF feature selection algorithm is used to select the optimal eigenvector and the support vector machine based on balanced decision tree is used to classify and recognize different dumbbell movements. Through testing on the dumbbell motion recognition system independently developed in the laboratory, the results show that the system can recognize the dumbbell movement within a single dumbbell movement cycle,and the recognition rate can reach more than 90%,which lays the foundation for providing more personalized dumbbell action guidance.