基于手机传感器识别行人步态的PSO-ELM算法

PSO-ELM algorithm for pedestrian gait recognition through mobile phone sensors

针对因手机携带位置不同对传感器产生干扰而导致行人步态识别准确率降低的问题,提出了一种粒子群优化极限学习机(PSO-ELM)识别算法。首先,基于极限学习机(ELM)分类方法,借助分层ELM多层降维的特点,利用粒子群优化算法对ELM算法参数进行寻优,设计有效识别行人手机携带位置的分层PSO-ELM分类方法。然后,通过线性判别分析的降维算法和PSO-ELM完成对行人步态的有效识别。实验使用Android手机对五种携带位置四种步态下的加速度和角速度数据进行采集,结果表明:在识别手机携带位置层面,训练集与测试集的识别准确率分别达到99.54%、99.47%;在识别行人步态层面,两种准确率分别达到95.74%、95.31%,证明所提算法具有较高的步态识别准确率。

Aiming at the problem that the accuracy of pedestrian gait recognition is reduced due to the interference of sensors caused by different positions of mobile phones,a particle swarm optimization-extreme learning machine(PSO-ELM)recognition algorithm is proposed.Firstly,based on the classification method of ELM and the characteristics of multi-level dimension reduction of hierarchical ELM,PSO algorithm is used to optimize the parameters of ELM model,and a hierarchical PSO-ELM classification method is designed to effectively identify the location of pedestrians'mobile phones.Then,through the dimensionality reduction algorithm of linear discriminant analysis and PSO-ELM,the effective recognition of pedestrian gait is completed.In the experiments,the acceleration and angular velocity data under four gaits of five carrying positions are collected by Android phones.The results show that the recognition accuracy of the training set and the test set are 99.54% and 99.47% at the level of identifying the carrying position of the mobile phone.At the level of identifying pedestrian gait,the accuracy of the two sets reach 95.74%and 95.31%,which proves that the proposed algorithm has high gait recognition accuracy.