普适计算环境下多模态信息融合的活动识别研究

发布时间：2018-03-13 08:46

本文选题：普适计算　切入点：活动识别　出处：《浙江大学》2016年硕士论文　论文类型：学位论文

【摘要】：有效融合多模态的情境信息是普适环境下活动识别的重要研究方向。而现有活动识别方法在普适环境下的可用性不高。一方面,传统的活动识别方法通常都有着两个潜在的假设前提：其一是传感器在活动识别中总是固定在人的身上,其二是传感器在活动识别中不会因环境的干扰而发生传感器的滑动、扭曲和旋转的情况。但在普适环境下的应用中,这些假设对参与人员的活动束缚较强,只能存在于可控的实验室环境下。另一方面,由于单一模态的运动信息(如加速度传感器信息)只能记录活动的部分信息,对复杂多变人类活动的识别效果不高。为了解决上述问题,提出了普适计算环境下多模态信息融合的活动识别方法。首先,该方法针对基于智能手机的活动识别中设备朝向与位置的影响,结合坐标系映射和多维时序数据分类的技术,实现了一种具有可解释性的和朝向、位置不变性活动识别方法。其次,该方法针对基于单一模态信息的活动识别效果不高的问题,结合模型间的多样性,实现了有效融合多模态信息的活动识别方法。实验结果表明,该方法在降低活动识别的设备朝向与位置影响上优于传统方法,能够有效融合多模态的信息提高活动识别的效果。
[Abstract]:The effective fusion of multi-modal situational information is an important research direction of activity recognition in pervasive environments, but the existing methods of activity recognition are not highly available in pervasive environments. The traditional method of activity recognition usually has two underlying hypotheses: one is that the sensor is always fixed on the human body in the activity recognition, the other is that the sensor will not slide because of the interference of the environment in the activity recognition. Distortion and rotation. But in general applications, these assumptions are more restrictive to the activity of participants and can only exist in controlled laboratory environments. Since the motion information of a single mode (such as acceleration sensor information) can only record part of the information of activity, the recognition effect of complex and variable human activities is not high. In order to solve the above problems, In this paper, an activity recognition method based on multi-modal information fusion in pervasive computing environment is proposed. Firstly, aiming at the influence of the orientation and position of the device in mobile recognition based on smart phone, the method combines coordinate system mapping and multi-dimensional temporal data classification technology. A method of activity recognition with interpretable and orientation and position invariance is implemented. Secondly, the method aims at the problem of low recognition effect based on single modal information and combines the diversity of models. The experimental results show that this method is superior to the traditional method in reducing the influence of the orientation and position of the equipment, and can effectively fuse the multi-modal information to improve the effect of activity recognition.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】