基于Wifi信号的手势识别技术研究

发布时间：2019-03-28 17:24

【摘要】：WiFi信号作为一种工作在2.4GHz和5.8GHz频段的无线电波,具有波长小,频率高,带宽足的特点,适合大量数据传输,因而在短距离无线通信领域的到广泛应用。随着模式识别和人机交互技术的不断发展,WiFi信号在目标探测和识别方面的强大能力被逐渐挖掘出来。如今,研究人员借助WiFi信号已经能够识别出目标的位置、人体的姿势,甚至是手势,基于WiFi信号的识别技术已然成为研究热点。基于以上背景,本文对利用WiFi信号实现手势识别的关键技术进行了探讨,构建出了初步的手势识别模型,并对其中涉及到的信号处理、特征提取和分类识别算法进行了深入的剖析。当WiFi信号在传播过程中遇到动态手势时,其幅度、相位和功率等传输特性会受到一定的影响,这种影响是由手势的移动特征决定的,这就意味着穿过手势的WiFi信号在某种意义上受到了手势的调制,包含了手势移动特性的信息,只要采用合理的方式将这一信息解调出来,就能够实现动作的识别。一般来说,完善的手势识别过程首先需要通过数据采集、数据预处理来建立适合特征提取的手势模型;其次采用特定的特征提取算法对手势进行特征提取以获取相应的特征向量;然后,构建能够对特征向量进行有效分类的识别算法模型;最后,为了验证识别方法的有效性,常常需要将特征向量分为训练集合和测试集合,训练集合作为分类识别算法的输入以训练识别模型,测试集合则输入训练好的识别模型以获取识别率,验证识别算法的有效性。在本文中,WiFi信号数据的采集由SORA软件无线电平台完成,本文保留802.11数据帧的长前导部分作为原始数据,通过数据预处理获取WiFi信号的功率包络,并对该包络进行周期分割,把得到的周期片段作为手势模型;为了降低原始样本中所包含信息的维度,同时减少无关噪声的干扰,本文采用离散小波变换(DTW)对原始样本进行特征提取,在保留足够的手势运动信息的条件下,大幅度压缩了特征数据量;在分类识别阶段,本文选择支持向量机(SVM)作为建立分类识别模型的主要算法,同时采用动态时间规整(DTW)算法对支持向量机的核函数进行了改造,以保证支持向量机能够适用于变长特征向量的分类。仿真和测试结果表明,本文提出的基于WiFi信号的手势识别技术的方法模型能够在少量样本条件下,有效识别9个预先定义的常用动态手势,平均识别率可达94.8%,具有一定的研究价值和实用性,为相关问题的解决提供了新的思路。
[Abstract]:As a kind of radio wave operating in 2.4GHz and 5.8GHz band, WiFi signal has the characteristics of small wavelength, high frequency and sufficient bandwidth, so it is suitable for a large number of data transmission, so it is widely used in the field of short-range wireless communication. With the development of pattern recognition and human-computer interaction technology, the powerful ability of WiFi signal in target detection and recognition is gradually excavated. Now, researchers have been able to identify the location of the target, the posture of the human body, and even the gesture with the help of the WiFi signal. The recognition technology based on the WiFi signal has become a hot research topic. Based on the above background, this paper discusses the key technology of using WiFi signal to realize gesture recognition, constructs a preliminary gesture recognition model, and processes the signal involved in it. The algorithms of feature extraction and classification recognition are deeply analyzed. When a WiFi signal encounters a dynamic gesture in the process of propagation, its transmission characteristics, such as amplitude, phase and power, will be affected to a certain extent, which is determined by the movement characteristics of the gesture. This means that the WiFi signal passing through the gesture is modulated by the gesture in a certain sense, which contains the information of the movement characteristic of the gesture. So long as the information is demodulated in a reasonable way, the motion recognition can be realized. Generally speaking, the perfect gesture recognition process first needs to establish the gesture model which is suitable for feature extraction through data acquisition and data pre-processing. Secondly, the special feature extraction algorithm is used to extract the gesture feature to obtain the corresponding feature vector, and then, the recognition algorithm model which can classify the feature vector effectively is constructed. Finally, in order to verify the effectiveness of the recognition method, it is often necessary to divide the feature vectors into training set and test set, and the training set is used as the input of the classification recognition algorithm to train the recognition model. The test sets input the trained recognition model to obtain the recognition rate and verify the effectiveness of the recognition algorithm. In this paper, the acquisition of WiFi signal data is accomplished by SORA software radio platform. In this paper, the long leading part of 802.11 data frame is retained as the original data, and the power envelope of WiFi signal is obtained by data preprocessing. The periodic segmentation of the envelope is carried out and the obtained periodic segment is used as the gesture model. In order to reduce the dimension of the information contained in the original sample and reduce the interference of irrelevant noise, the discrete wavelet transform (DTW) is used to extract the features of the original sample. The feature data is greatly compressed; In the stage of classification and recognition, support vector machine (SVM) is chosen as the main algorithm to establish the classification and recognition model. Meanwhile, the kernel function of SVM is modified by dynamic time warping (DTW) algorithm. In order to ensure that support vector machine can be applied to variable length feature vector classification. The simulation and test results show that the proposed method model of gesture recognition based on WiFi signal can effectively recognize 9 pre-defined dynamic gestures with an average recognition rate of 94.8% under a small number of samples. It has certain research value and practicability, and provides a new way to solve the related problems.
【学位授予单位】：北京邮电大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41;TN92

【参考文献】