基于层次分类的手机位置无关的动作识别

发布时间：2018-12-17 10:20

【摘要】：使用智能手机中集成的加速度传感器识别用户日常动作在惯性定位、个性化推荐、运动量评估等领域有重要的应用。手机位置不固定导致的动作识别率低下是该领域面临的主要问题。为了提高手机位置不固定时的动作识别率,该文提出一种基于层次分类的动作识别方法。该方法将动作识别分为多层,每一层包含一个分类器。在训练某一层分类器时,首先根据本层训练样本集进行特征选择并训练分类器。然后使用训练得到的分类器对训练样本分类,并计算分类结果的可信度。最后通过对低可信度的样本进行剪枝得到下层分类器的训练样本。对未知类别的样本分类时,首先使用第1层分类器分类。如果分类结果可信度较高,则分类结束;否则使用下层分类器分类,直至所有分类器遍历完。实验部分通过对采集的动作数据进行仿真,验证了该文方法的有效性。结果表明,与单层分类器相比,该方法可以将动作识别率由85.2%提高至89.2%。
[Abstract]:Using the accelerometer integrated in the smart phone to identify the user's daily action has important applications in the fields of inertial positioning, personalized recommendation, motion quantity evaluation and so on. The low recognition rate caused by mobile phone location is the main problem in this field. In order to improve the recognition rate of mobile phone when its position is not fixed, this paper presents an action recognition method based on hierarchical classification. In this method, action recognition is divided into multiple layers, each layer contains a classifier. In training a layer classifier, the feature selection and classifier are firstly carried out according to the training sample set of this layer. Then the training samples are classified using the training classifier and the reliability of the classification results is calculated. Finally, the training samples of the lower classifier are obtained by pruning the samples with low credibility. When classifying samples of unknown categories, the first layer classifier is used. If the classification result is reliable, the classification ends; otherwise, the lower classifier is used until all the classifiers have been traversed. In the experiment part, the validity of the method is verified by the simulation of the collected action data. The results show that the method can improve the recognition rate from 85.2% to 89.2% compared with the single-layer classifier.
【作者单位】：南开大学计算机与控制工程学院;
【基金】：天津市重大科技专项(13ZCZDGX01098) 天津市自然科学基金(16JCQNJC00700)~~
【分类号】：TP212.9

【相似文献】