WO2016197648A1 - 一种基于无线信号的动作检测和识别的方法 - Google Patents

一种基于无线信号的动作检测和识别的方法 Download PDF

Info

Publication number
WO2016197648A1
WO2016197648A1 PCT/CN2016/076575 CN2016076575W WO2016197648A1 WO 2016197648 A1 WO2016197648 A1 WO 2016197648A1 CN 2016076575 W CN2016076575 W CN 2016076575W WO 2016197648 A1 WO2016197648 A1 WO 2016197648A1
Authority
WO
WIPO (PCT)
Prior art keywords
wireless
signal
data
motion
signal strength
Prior art date
Application number
PCT/CN2016/076575
Other languages
English (en)
French (fr)
Inventor
刘向阳
王玮
Original Assignee
南京大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 南京大学 filed Critical 南京大学
Publication of WO2016197648A1 publication Critical patent/WO2016197648A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/02—Preprocessing
    • G06F2218/04—Denoising
    • G06F2218/06—Denoising by applying a scale-space analysis, e.g. using wavelet analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
    • G06F2218/08—Feature extraction

Definitions

  • the present invention relates to the field of wireless networks and pervasive computing, and more particularly to a method for motion detection and recognition based on wireless signals.
  • Human motion detection and recognition technology is a core technology of ubiquitous computing. With the rapid development of industries such as wearable computing and smart home, in recent years, applications based on various behavior recognition technologies have shown a situation of spurt development. For example, motion monitoring and recording technology based on mobile phones and wristbands can help consumers understand their own sports and sleep behaviors, thereby actively changing their lifestyle. The elderly care system with human motion monitoring function can provide applications such as fall alarms and life routine detection, which are very important for elderly care. In the intelligent security system, behavior recognition technology can also be used to determine whether abnormal behavior occurs in the monitored area.
  • wireless access devices have spread to thousands of households and various public places. Since the human body is a good conductor of electricity, people have a strong reflection effect on radio waves. In this way, wireless devices that can be seen everywhere around us can actually play the role of "human radar.”
  • the key advantages of motion detection and recognition technology based on wireless signals are: First, the monitored person does not need to wear any equipment, so the non-active matching target can be monitored; in addition, the system only needs to use existing common wireless devices, such as notebooks. , mobile phones, WiFi routers, etc., software upgrades can be achieved, the cost is very low.
  • the existing human motion detection system based on the universal wireless device has weak performance and can only recognize one or two actions, and the recognition effect is affected by the surrounding environment.
  • CN103606248A He proposed to use support vector machine and anomaly detection technology to achieve the judgment of falling.
  • the limitation is that the system can only recognize the action of falling, and the system detection distance is limited.
  • the present invention has been made to improve the above technical drawbacks.
  • Excessive noise in the measurement results of general-purpose wireless devices is a fundamental problem that limits the effectiveness of motion detection and identification.
  • the invention proposes a method of principal component analysis to jointly process multi-channel wireless signal data, and proposes components related to human motion, thereby achieving the effect of eliminating noise.
  • the present invention acquires the speed characteristics of the motion by the influence of human motion on the fluctuation of the wireless signal, and performs motion recognition. This method reduces the impact of the environment on the signal, so that it can obtain better recognition results in different environments.
  • the object of the present invention is to propose a motion detection and motion recognition method based on wireless signals, aiming at the weakness of the existing wireless signal motion recognition technology, solving how to obtain a reliable wireless signal through a universal wireless device, and achieving a comparison in different environments. High recognition efficiency issues.
  • the present invention is implemented by the following technical solution: a motion detection and motion recognition method based on wireless signals, characterized in that wireless signal data is collected by using one or more general wireless devices, and multiple wireless signals are transmitted. Correlation of data Denoising the wireless data, and extracting the characteristics related to the speed of the human body from the wireless data to detect and identify the motion, including the following basic steps:
  • Data acquisition by measuring the received wireless signal, including measuring the RSSI and CSI of each data packet;
  • PCA Principal component analysis
  • OFDM Orthogonal Frequency Division Multiplexing
  • Motion feature extraction that is, using time-frequency analysis of wireless signal data to obtain the intensity of the wireless signal at different frequencies, and the time-frequency analysis method includes Short-Time Fourier Transform (STFT) and Wavelet Transform (Wavelet Transform);
  • STFT Short-Time Fourier Transform
  • Wavelet Transform Wavelet Transform
  • Model training that is, the training of the offline data set by the system, the training data set is established by collecting the wireless signals corresponding to different actions, and the signals are manually labeled and segmented during the collection process, and the labeling process indicates which specific signals belong to which An action, the segmentation is manually marking the start and end points of the action, and for each action, collecting signals of different people in different environments to form a training data set, for the characteristics of an action instance, adopting
  • the method including vector quantization first quantizes the signal strength of each frame or directly uses a mixed Gaussian hidden Markov model for training;
  • Motion recognition and motion recognition method adopt Hidden Markov Model (HMM), which inputs each frame signal strength vector of motion feature extraction into multiple hidden Markov models, and each hidden Markov model corresponds to one action. For each hidden Markov model, the system calculates the probability of generating a current signal strength vector sequence, and the system selects the action corresponding to the model with the largest possible generation of the current signal vector sequence as the recognition result.
  • HMM Hidden Markov Model
  • the universal wireless device refers to a wireless device supporting WiFi, Long Term Evolution (LTE), Bluetooth, and Zigbee communication technologies.
  • the wireless signal data includes a Received Signal Strength Indication (RSSI) and a Channel State Indication (CSI).
  • RSSI Received Signal Strength Indication
  • CSI Channel State Indication
  • the beneficial effects of the present invention are as follows: aiming at the defects that the existing human motion recognition technology cannot be implemented on the general wireless device, a method for performing noise reduction processing on the wireless signal data by using multiple signals is proposed, and the action is performed by using a relatively stable motion speed feature. Identify.
  • the benefit is that motion recognition can be implemented on existing wireless devices with a simple software upgrade.
  • the system can achieve better recognition results in different types of environments, such as indoor line of sight, indoor no line of sight and outdoor.
  • the anti-jamming capability of the wireless multipath is also stronger by utilizing the characteristics of the motion speed.
  • Figure 1 is an application scenario of the present invention.
  • Figure 2 is a flow chart showing an embodiment of the present invention.
  • Figure 3 is an embodiment of a denoising process.
  • 4 is a diagram showing an example of collected wireless signal strength raw data.
  • FIG. 5 is a diagram showing an example of wireless signal strength after denoising processing.
  • Figure 6 is a diagram showing an example of signal intensity variation caused by walking.
  • Fig. 7 is a diagram showing an example of changes in signal intensity caused by sitting down.
  • Fig. 8 is a view showing an example of a change in signal intensity caused by a fall.
  • Fig. 9 is a diagram showing an example of a feature acquired from a fall signal.
  • the invention utilizes the interference of the human body motion on the wireless signal, uses the universal wireless device to collect the wireless signal data, denoises the data, and extracts the feature associated with the motion speed to identify the action.
  • Collecting wireless signal data using one or more general-purpose wireless devices denoising wireless data through correlation of multiple wireless signal data, and extracting features related to human motion speed from wireless data to detect and identify motion .
  • the main technical features are: general wireless device; second, denoising processing by correlation; third, moving speed characteristics.
  • the basic steps include: data acquisition, data denoising, data segmentation, feature extraction, model training, and motion recognition.
  • the denoising process performs principal component analysis (PCA) on the multiplex signal to obtain the data component with the lowest noise.
  • PCA principal component analysis
  • the multiplex signal refers to measurement data on different subcarriers in Orthogonal Frequency Division Multiplexing (OFDM), and/or measurement data of different transmit/receive antennas, and/or different Measurement data between devices.
  • OFDM Orthogonal Frequency Division Multiplexing
  • the universal wireless device refers to a wireless device supporting WiFi, Long Term Evolution (LTE), Bluetooth, and Zigbee communication technologies.
  • the wireless signal data refers to a Received Signal Strength Indication (RSSI) and/or a Channel State Indication (CSI).
  • RSSI Received Signal Strength Indication
  • CSI Channel State Indication
  • the motion feature extraction uses time-frequency analysis of wireless signal data to obtain wireless The strength of the signal at different frequencies.
  • time-frequency analysis refers to a short-time Fourier transform (STFT) and/or a wavelet transform (Wavelet transform).
  • STFT short-time Fourier transform
  • Wavelet transform wavelet transform
  • the motion recognition method adopts a Hidden Markov Model (HMM).
  • HMM Hidden Markov Model
  • FIG. 1 is an embodiment of the present invention.
  • One or more wireless transmitters 101 are included in the scene.
  • the wireless transmitter 101 can employ WiFi, LTE, Bluetooth or ZigBee technology, and the signal transmitted by the wireless transmitter 101 is a normal data message conforming to the corresponding wireless technology protocol.
  • One or more wireless receivers 102 are included in the scene.
  • the wireless receiver 102 receives the signals of the wireless transmitter.
  • the wireless receiver 102 can measure the strength of the wireless signal transmitted by the wireless transmitter 101 by RSSI or CSI.
  • RSSI or CSI As shown in FIG. 1, from the wireless transmitter 101 to the wireless receiver 102, radio waves can be reached through two different paths.
  • the line-of-sight (LOS) path 104 is a direct path, and the reflection path 105 is reflected by the moving body 103 to reach the wireless receiver 102.
  • LOS line-of-sight
  • the reflection path 105 is constantly changing, thereby causing a change in the signal strength received by the wireless receiver 102.
  • the present invention detects and recognizes human motion by the intensity changes received by the wireless receiver 102.
  • the invention is characterized in that the object 103 to be detected does not need to be worn with any special equipment or sensors, and its action can be recognized completely by reflection from the human body.
  • the wireless transmitter 101 and the wireless receiver 102 may be general-purpose electronic devices including, but not limited to, mobile phones, wireless routers, cellular base stations, notebook computers, smart set top boxes, wireless sensors, smart wearable devices, and the like.
  • the above general electronic device only needs to have a wireless signal transceiver to be used in the present invention, and no special hardware modification is required.
  • wireless signal data acquisition 201 is first performed, and then the data is subjected to denoising processing 202.
  • the denoised processed data can be used for offline model generation or for online motion recognition.
  • the motion is first detected and segmented 203, then the feature extraction 204 is performed, and finally the model-based motion recognition 206 is performed using the extracted features in combination with the motion model.
  • the model of the comparison in step 206 is obtained by offline model training 205. It is worth noting that the above process is only one possible implementation of the present invention, human motion recognition and detection.
  • the measurement system can also be implemented in a variety of other ways.
  • the wireless signal data acquisition employed by the present invention is implemented by the wireless receiver 102.
  • the wireless receiver 102 performs intensity detection on the signals of the wireless transmitter 101. This can be accomplished by the wireless transmitter 101 transmitting data at a particular rate, such as 2500 packets per second, while the wireless receiver 102 performs intensity measurements on each data packet. In addition, measurements can also be made by intensity detection of the daily data flow of the wireless transmitter 101.
  • the specific data acquisition method can be implemented by measuring the RSSI or CSI of each data packet. Most popular wireless devices currently support wireless measurement data via RSSI or CSI. Of course, we do not exclude the wireless signal receiver 102 from measuring the received wireless signal by other means.
  • Wireless signal measurements can be multiplexed.
  • the multipath here refers to the following situations: First, when the wireless signal adopts OFDM modulation, the signal strength on multiple subcarriers can be measured separately; second, when the wireless transmitter 101 or the wireless receiver 102 has multiple antennas, each The signal strength on the transmit/receive antenna pair can be measured separately; third, when there are multiple wireless transmitters or multiple wireless receivers, the signal strength on each pair of transmitters/receivers can be measured separately. These individually measured wireless data can be viewed as separate multiplexed signals.
  • the invention proposes a joint denoising algorithm for multi-path signals, which is mainly based on the correlation of human motion in multi-path signals. Using this correlation, principal component analysis can be used to extract motion-related information from multiple signals.
  • FIG. 3 illustrates a specific method using principal component analysis by taking multiple signals on multiple subcarriers in OFDM as an example.
  • the wireless receiver 102 can measure a series of signal strengths on each subcarrier. We first arrange the signal strengths 301 on different subcarriers in order of time and subcarriers. We then preprocess the sequence of signal strengths on a single subcarrier, which subtracts the long-term average of the signal strength from the signal strength time series to obtain the pre-processed signal strength 302.
  • the pre-processed signal strength is segmented to obtain a measurement data matrix H.
  • Each row of the measurement matrix H represents the signal strength on a single subcarrier, at the same time on each column The measured signal strength on different subcarriers.
  • the number n of rows of the measurement matrix H is equal to the number of subcarriers, and the number of columns m of the measurement matrix is equal to the length of the time series.
  • the dimension of the measurement matrix H is 30 x 2500.
  • the dimension of the correlation matrix 303 is n x n.
  • the feature vector 304 is arranged according to the size of the feature value, and is represented as q 1 , q 2 , . . . , q n . Each feature vector has a length of n.
  • the measurement matrix H is multiplied by the feature vector 304 to obtain the individual PCA components of the signal.
  • PCA component 305 is a relinear combination of multiplexed signals. Therefore, the noise in each signal is weakened due to its uncorrelated characteristics. On the other hand, the interrelated human motion information in each signal will be enhanced in the first few components of the PCA. Normally, we take the second component of PCA as a result of denoising.
  • Figure 4 and Figure 5 plot the original acquired CSI signal strength and signal strength after denoising. We can see that the signal noise after processing is greatly reduced. At the same time, the high frequency components of the signal are still preserved.
  • the online recognition system After denoising, the online recognition system first needs to detect and segment the action.
  • the variance information of the signal can be utilized to determine whether there is an action.
  • the wireless signal strength will fluctuate significantly, and the judgment can be made by increasing the variance.
  • the determination may be made, for example, by the smoothness of the feature vector of the PCA described above.
  • the feature vector tends to be smooth, the correlation of each signal is enhanced, and an action may occur at this time.
  • the action can be split.
  • specific actions are identified based on the segmented action information.
  • the specific segmentation method can use a fixed time segmentation, such as cutting into a segment every 3 seconds. Or split according to the start and end time of the action.
  • the present invention employs features related to the speed of movement of the human body for motion recognition.
  • the speed characteristics of different movements of the human body are different. For example, when walking, the overall movement speed of the human body is about 1 m / s, while the running speed can reach 2-3 m / sec. In addition to this, the speed of the movement changes regularly. For example, when a person falls, there is a process of accelerating the free fall, and the speed of movement is significantly accelerated. Therefore, the movement speed information can be used to identify various actions of the person, including: walking, running, brushing, pushing hands, standing up/sitting, falling, opening/closing, boxing, and the like.
  • the change of wireless signal strength is affected by human motion, and the frequency of change is related to the speed of human motion.
  • a wireless signal having a carrier frequency of 5 GHz has a wavelength of 6 cm.
  • the wireless signal strength will vary at a frequency of 33 Hz corresponding to the motion speed of 1 m/s.
  • Figures 6, 7, and 8 illustrate examples of changes in wireless signal strength caused by walking, sitting, and falling, respectively. It can be clearly seen from the figure that the speed of walking is higher than that of sitting, and in the process of falling, there is a signal frequency that is accelerated, that is, the process of motion acceleration.
  • the intensity variation frequency of the wireless signal can be obtained by various time-frequency analysis methods.
  • Common time-frequency analysis methods include short-time Fourier transform and wavelet transform.
  • the short-time Fourier transform divides the signal into frames by means of windowing, calculates the Fourier transform value of each frame signal, and obtains the strength of the signal at different frequencies.
  • each frame of the signal will produce an intensity vector, each of which represents the strength of the signal at a certain frequency in the frame.
  • a typical frame length can take 512 or 1024 sample points, and 32 or 64 sample points can be moved between frames. This allows you to get the difference in signal strength over time and frequency.
  • the wavelet transform can also be used to obtain the signal strength vector of each frame signal in each frequency band.
  • Figure 9 is an example of features acquired from a fall signal.
  • the abscissa is time, the ordinate is frequency, and the brightness of the square represents the energy of the signal.
  • the signal energy is concentrated on The low frequency part indicates that the movement speed is low.
  • the time-frequency energy characteristics of the signal can be extracted by a short-time Fourier transform or a wavelet transform. Using this energy feature, pattern recognition can be used to identify specific actions.
  • One such implementation is to use a hidden Markov model for identification.
  • the hidden Markov model is widely used in speech signal recognition, and its application in wireless signal motion recognition is similar to that in speech signal recognition.
  • the specific implementation manner may establish a training data set by collecting wireless signals corresponding to different actions. Signals can be manually labeled and segmented during the acquisition process. The labeling process is to indicate which action a particular signal belongs to. Segmentation is the manual marking of the start and end points of the action. For each action, a plurality of different people can be collected, and signals of actions are performed in different place environments to form a training data set. For each specific action in the training data set, such as walking, the feature extraction method described above can be used to acquire features.
  • the vector quantization method can be used to quantize the signal strength of each frame, or directly by using the mixed Gaussian hidden Markov model.
  • the traditional expectation maximization algorithm can be used to iteratively generate the action model.
  • Offline training generates a corresponding hidden Markov model for each specific action. This model can be used for online motion recognition.
  • the online recognition system After performing motion detection and feature extraction, the online recognition system inputs the extracted frame signal strength vectors into multiple hidden Markov models. Each hidden Markov model corresponds to an action. For each Hidden Markov Model, the system calculates its likelihood of generating a current signal strength vector sequence. The system selects an action corresponding to the model having the largest possible generation of the current signal vector sequence as the recognition result.

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mobile Radio Communication Systems (AREA)

Abstract

一种基于无线信号的动作检测和识别的方法，涉及无线网络和普适计算领域，尤其涉及一种基于无线信号的动作检测和识别的方法。该方法利用人体运动对无线信号的干扰，采用通用无线设备采集无线信号数据（201），对数据进行去噪处理后提取与运动速度相关联的特征对动作进行识别。使用一个或多个通用无线设备采集无线信号数据（201），通过多路无线信号数据的相关性对无线数据进行去噪处理（202），并从无线数据中提取出与人体运动速度相关的特征对运动进行检测和识别。基本步骤包括：数据采集（201）、数据去噪（202）、数据分段（203）、特征提取（204）、模型训练（205）、动作识别（206）。

Description

一种基于无线信号的动作检测和识别的方法 技术领域
本发明涉及无线网络和普适计算领域，尤其涉及一种基于无线信号的动作检测和识别的方法。
背景技术
人体动作检测和识别技术是普适计算的一项核心技术。随着可穿戴计算、智能家居等产业的迅速发展，近年来基于各类行为识别技术的应用呈现出井喷式发展的局面。例如，基于手机和手环的运动监测和记录技术可以帮助消费者了解自己的运动和睡眠行为，从而主动地改变自己的生活习惯。具有人体动作监测功能的老人看护系统可以提供跌倒报警、生活规律检测等对老人看护来说非常关键的应用。在智能安防系统中，也可以通过行为识别技术判断是否在监控区域内有非正常行为发生。
虽然人体行为识别技术非常有用，但制约其发展的一个重要因素是其成本及便利性。传统的行为识别技术一般需要通过手机、手环、传感器等专用设备来采集用户信息。一方面，用户需要长期佩戴这些设备，给用户生活带来一定的不便。另一方面，用户需要专门为某个应用购买和安装特定的传感设备，成本较高。
随着Wi-Fi，3G等无线技术的发展，无线接入设备已经普及到千家万户以及各类公共场所。由于人体是电的良导体，人对无线电波有较强的反射作用。这样，在我们身边随处可见的无线设备实际可以起到“人体雷达”的作用。基于无线信号的动作检测和识别技术的关键优势在于：首先，被监测人不需要佩戴任何设备，所以可以对非主动配合目标进行监测；此外，系统仅需要对现有常见的无线设备，如笔记本、手机、WiFi路由器等，进行软件升级就可以实现监测，成本非常低廉。
现有的基于通用无线设备的人体动作检测系统性能较弱，只能对一到两个动作进行识别，且识别效果受到周边环境影响。例如，在CN103606248A中伍 楷舜等提出使用支持向量机和异常检测技术来实现对摔倒的判断。其局限在于系统只能对摔倒一种动作进行识别，并且系统检测距离有限。
本发明对以上的技术缺陷进行了改进。通用无线设备测量结果的噪声过大是限制其动作检测和识别有效性的根本问题。本发明提出利用主分量分析的方法，对多路无线信号数据进行联合处理，从中提出与人体动作相关的分量，从而达到消除噪声的效果。另一方面，本发明通过人体运动对无线信号波动的影响，获取动作的速度特征，进行动作识别。这种方法降低了环境对信号的影响，从而可以在不同环境下均获取较好的识别效果。
发明内容
本发明的目的是：提出一种基于无线信号的动作检测和动作识别方法，针对现有无线信号动作识别技术的弱点，解决如何通过通用无线设备获取可靠的无线信号，并在不同环境下达到较高识别效率的问题。
具体地说，本发明是采用以下的技术方案来实现的：一种基于无线信号的动作检测和动作识别方法，其特征在于，使用一个以上的通用无线设备采集无线信号数据，通过多路无线信号数据的相关性对无线数据进行去噪处理，并从无线数据中提取出与人体运动速度相关的特征对运动进行检测和识别，包括如下基本步骤：
数据采集，通过对接受到的无线信号进行测量包括测量每个数据包的RSSI、CSI；
数据去噪，对多路信号进行主成分分析(PCA,Principal component analysis)，获取噪声最低的数据成分，多路信号包括正交频分复用(OFDM,Orthogonal Frequency Division Multiplexing)中不同子载波上的测量数据、不同发射或接收天线的测量数据、不同设备之间的测量数据；
运动特征提取，即采用对无线信号数据进行时频分析获取无线信号在不同频率上的强度，时频分析方法包括短时傅立叶变换(STFT，Short-time Fourier transformation)和小波变换(Wavelet transform)；
模型训练，即系统对离线数据集的训练，通过采集不同动作对应的无线信号来建立训练数据集，在采集过程中对信号进行人工的标注和分段，标注过程是标明特定的信号是属于哪一种动作，分段是人工将动作开始和结束点标出，针对每个动作，采集多个不同人在不同地点环境下进行动作的信号，形成训练数据集，针对一个动作实例的特征，采用包括矢量量化的方法先对每帧信号强度进行量化或者直接利用混合高斯隐马尔可夫模型进行训练；
动作识别，动作识别方法采用隐马尔可夫模型(HMM，Hidden Markov Model)，即将动作特征提取的各帧信号强度矢量分别输入多个隐马尔可夫模型，每个隐马尔可夫模型对应一个动作，对每一个隐马尔可夫模型，系统计算其产生当前信号强度矢量序列的可能性，系统选择具有最大可能生成当前信号矢量序列的模型对应的动作作为识别结果。
上述技术方案的进一步特征在于，所述通用无线设备是指支持WiFi，长期演进(LTE,Long Term Evolution)，蓝牙(Bluetooth)，Zigbee通信技术的无线设备。
上述技术方案的进一步特征在于，所述无线信号数据包括接收信号强度指示(RSSI，Received signal strength indication)和信道状态指示(CSI，Channel state indication)。
本发明的有益效果如下：针对现有人体动作识别技术无法在通用无线设备上实现的缺陷，提出通过多路信号对无线信号数据进行降噪处理的方法，并利用较为稳定的运动速度特征对动作进行识别。其有益效果在于，可以通过简单的软件升级在现有无线设备上实现动作识别。除此之外，系统可以在不同类型的环境下，如，室内视距、室内无视距以及室外，均达到较好的识别效果。利用运动速度的特征对无线多径的抗干扰能力也更强。
附图说明
图1是本发明的一个应用场景。
图2是本发明的一种实施流程图。
图3是去噪处理的一个实施例。
图4是采集到的无线信号强度原始数据的示例图。
图5是经过去噪处理后的无线信号强度示例图。
图6是走路造成的信号强度变化示例图。
图7是坐下造成的信号强度变化示例图。
图8是摔倒造成的信号强度变化示例图。
图9是从摔倒信号中获取的特征示例图。
具体实施方式
以下结合附图和具体实施例对本发明作进一步详细说明。
本发明利用人体运动对无线信号的干扰，采用通用无线设备采集无线信号数据，对数据进行去噪处理后提取与运动速度相关联的特征对动作进行识别。
使用一个或多个通用无线设备采集无线信号数据，通过多路无线信号数据的相关性对无线数据进行去噪处理，并从无线数据中提取出与人体运动速度相关的特征对运动进行检测和识别。主要技术特征是一、通用无线设备；二、利用相关性进行去噪处理；三、运动速度特征。基本步骤包括：数据采集、数据去噪、数据分段、特征提取、模型训练、动作识别。
进一步地，所述去噪处理对多路信号进行主成分分析(PCA,Principal component analysis)，获取噪声最低的数据成分。
进一步地，所述多路信号指的是正交频分复用(OFDM,Orthogonal Frequency Division Multiplexing)中不同子载波上的测量数据，与/或不同发射/接收天线的测量数据，与/或不同设备之间的测量数据。
进一步地，所述通用无线设备指的是支持WiFi，长期演进(LTE,Long Term Evolution)，蓝牙(Bluetooth)，Zigbee通信技术的无线设备。
进一步地，所述无线信号数据指的是接收信号强度指示(RSSI，Received signal strength indication)与/或信道状态指示(CSI，Channel state indication)。
进一步地，所述运动特征提取采用对无线信号数据进行时频分析获取无线 信号在不同频率上的强度。
进一步地，所述时频分析指短时傅立叶变换(STFT，Short-time Fourier transformation)与/或小波变换(Wavelet transform)。
进一步地，所述动作识别方法采用隐马尔可夫模型(HMM，Hidden Markov Model)。
图1是本发明的一个实施场景。场景中包含一个或多个无线发射器101。无线发射器101可以采用基于WiFi，LTE，蓝牙或ZigBee技术，无线发射器101发射的信号是符合相应无线技术协议的正常数据报文。场景中包含一个或多个无线接收器102。无线接收器102接收无线发射器的信号。无线接收器102可以通过RSSI或者CSI测量无线发射器101发送的无线信号的强度。如图1所示，从无线发射器101到无线接收器102，无线电波可以通过两条不同的路径到达。其中，视距(LOS，Line-of-sight)路径104是直达路径，反射路径105是通过运动中的人体103反射才到达无线接收器102的。因此，在人体103运动的条件下，反射路径105会不断变化，从而造成无线接收器102接收到的信号强度变化。本发明通过无线接收器102收到信号的强度变化来检测和识别人体运动。
本发明的特点在于，被检测的对象103不需要佩戴任何特殊设备或传感器，完全通过人体反射即可对其动作进行识别。无线发射器101和无线接收器102可以是通用的电子设备，包括但不限于：手机，无线路由器，蜂窝网基站，笔记本电脑，智能机顶盒，无线传感器和智能穿戴设备等。以上通用电子设备只需要具有无线信号收发器即可用于本发明，不需要进行特殊的硬件改造。
图2是本发明的一种实施流程。实施过程中首先进行无线信号数据采集201，随后对数据进行去噪处理202。去噪处理后的数据可以用于离线的模型生成，也可以用于在线的动作识别。在线动作识别过程中，首先要对动作进行检测和分段203，然后进行特征提取204，最后利用提取的特征结合动作模型进行基于模型的动作识别206。在步骤206中比对的模型是通过离线的模型训练205获取的。值得注意的是，以上流程只是本发明的一种可能实现方式，人体动作识别和检 测系统还可以通过其他各种方式来实现。
本发明所采用的无线信号数据采集是通过无线接收器102实现的。在采集过程中，无线接收器102对无线发射器101的信号进行强度检测。这可以由无线发射器101以特定速率，如，2500数据包每秒，来发送数据，同时无线接收器102对每个数据包进行强度测量来实现。除此之外，测量也可以通过无线发射器101日常数据流量进行强度检测实现。具体的数据获取方式可以通过测量每个数据包的RSSI或者CSI来实现。现有多数通用无线设备均支持通过RSSI或CSI提供无线测量数据。当然，我们也不排除无线信号接收器102通过其他方式对接收到的无线信号进行测量。
无线信号测量结果可以是多路的。这里的多路是指以下几种情况：一，无线信号采用OFDM调制时，多个子载波上的信号强度可以单独测量；二，当无线发射器101或无线接收器102拥有多根天线时，每对发射/接收天线对上的信号强度可以单独测量；三，当有多个无线发射器或多个无线接收器时，每对发射/接收器上的信号强度可以单独测量。这些单独测量的无线数据均可看成是独立的多路信号。
由于通用设备采集的无线信号数据通常含有较高的噪音，在进行进一步处理前需要进行去噪处理。本发明提出针对多路信号的联合去噪算法，其主要依据为人体运动在多路信号中的相关性。利用该相关性，可以使用主成分分析的方法来从多路信号中提取出与运动相关的信息。
图3以OFDM中多子载波上的多路信号为例说明了使用主成分分析的具体方法。在每一个子载波上无线接收器102都可以测量一系列的信号强度，我们首先将不同子载波上的信号强度301按照时间和子载波的顺序排列。随后我们对单个子载波上的信号强度序列进行预处理，预处理中将信号强度时间序列减去信号强度的长期平均值，以得到预处理后的信号强度302。
在此之后，将预处理后的信号强度进行分段处理，获取测量数据矩阵H。测量矩阵H的每一行表示的是单个子载波上的信号强度，每一列上为同一时刻 测量的不同子载波上的信号强度。测量矩阵H的行数n等于子载波的数量，测量矩阵的列数m等于时间序列的长度。例如，可以选择截取每秒钟的测量数据形成测量矩阵H。这时，如果每秒有2500个数据，则m＝2500。如果系统在30个OFDM上采集数据，则n＝30。测量矩阵H的维度就是30×2500。
对测量矩阵H进行相关操作可以获取其相关矩阵303，C＝H×HT，其中HT为H的转置矩阵。相关矩阵303的维度是n×n。对相关矩阵C进行奇异值分解(SVD Singular value decomposition)或特征值分解(eigendecomposition)，我们可以获取相关矩阵的特征值和特征向量304。其中，特征向量304按照特征值的大小排列，表示为q1，q2,…,qn。每一个特征向量的长度均为n。
测量矩阵H与特征向量304相乘可以获取信号的各个PCA成分。具体地，PCA成分hi可以使用公式hi＝qi×H来获取。PCA成分305是对多路信号的重新线性组合。所以，在各路信号中的噪声由于其互不相关的特点而被削弱。另一方面，各路信号中相互相关的人体动作信息将在PCA的前几个成分中得到加强。通常情况下，我们取PCA的第二成分来作为去噪后的结果。
图4和图5分别绘制了原始采集到的CSI信号强度和经过去噪处理后的信号强度。我们可以看出，处理后的信号噪声大大减弱。同时，信号的高频分量仍然得到了保留。
经过去噪处理后，在线识别系统首先需要对动作进行检测和分段。对动作进行检测可以采用多种方式。具体地，可以利用信号的方差信息来判断是否有动作。当有动作发生时，无线信号强度将有明显波动，此时可以通过其方差变大来进行判断。除此之外，还可以例如上述PCA的特征向量的平滑度来进行判断。当特征向量趋向平滑，说明各路信号的相关性增强，此时可能会有动作发生。当然也可以结合以上多种判断方式来进行综合判断。
当判断有动作发生后，可以对动作进行切分。在识别时依据分段后的动作信息来识别具体动作。具体切分方式可以使用固定时长的切分，如每隔3秒切为一段。或者根据动作开始和结束时间来进行切分。
识别具体动作首先需要提取动作特征。本发明采用与人体动作速度相关的特征来进行动作识别。人体的不同动作的速度特征是不同的。例如，走路时，人体的整体运动速度大约为1米/秒左右，而跑步时速度可以达到2-3米/秒。除此之外，动作的速度变化也是有规律的。例如，人在摔倒时会有一个自由落体的加速过程，运动速度明显加快。因此，通过运动速度信息可以对人的多种动作进行识别，具体包括：走路、跑步、刷牙、推手、站起/坐下、摔倒、开/关门、拳击等等。
无线信号强度变化受到人体运动的影响，其变化频率与人体动作速度有相关性。例如，载波频率为5GHz的无线信号的波长为6厘米。这样，当人体运动造成反射路径105的长度变化6厘米时，因干涉效应可以观察到无线信号强度由强变弱再变强的一个周期。因此，对应运动速度1米/秒的动作，无线信号强度将以33Hz的频率变化。测量无线信号强度变化的频率即可知道周围人体运动的速度。
图6、图7、图8分别绘制了走路、坐下和摔倒造成的无线信号强度变化的示例。从图中可以明显看出走路的运动速度高于坐下，而摔倒的过程中有一个信号频率加快，即运动加速的过程。
无线信号的强度变化频率可以采用多种时频分析方法来获取。常见的时频分析方法包括短时傅立叶变换和小波变换等等。短时傅立叶变换通过对信号加窗的方式，将信号分帧，计算每一帧信号的傅立叶变换值，获取信号在不同频率上的强度。这样，每一帧信号将产生一个强度矢量，矢量中每一个数据表示信号在本帧中的某个频率上的强度。典型的帧长可以取512或1024个采样点，帧和帧之间可以移动32或64个采样点。这样就可以获取信号强度在时间和频率上的分别。同样的，也可以利用小波变换获取每一帧信号在各个频段上的信号强度矢量。
图9是从摔倒信号中获取的特征的一个示例。横坐标为时间，纵坐标为频率，方块的亮度表示信号的能量。从图中可见，摔倒开始时，信号能量集中在 低频部分，说明运动速度较低。而1到1.5秒之间运动有一个迅速加速的过程。这主要表现在信号能量向高频部分移动，说明运动在加速。随后信号能量又回复到低频，说明运动停止。
通过短时傅立叶变换或小波变换，可以提取信号的时频能量特征。利用该能量特征就可以采用模式识别的方式来识别具体的动作。其中一种实施方式是采用隐马尔可夫模型来进行识别。
隐马尔可夫模型在语音信号识别中得到广泛应用，在无线信号动作识别中的应用方式与语音信号识别中类似。首先，系统需要进行离线数据集的训练。具体的实施方式可以通过采集不同动作对应的无线信号来建立训练数据集。在采集过程中，可以对信号进行人工的标注和分段。标注过程是标明特定的信号是属于那一种动作。分段是人工将动作开始和结束点标出。针对每个动作，可以采集多个不同人，在不同地点环境下进行动作的信号，形成训练数据集。对于训练数据集中的每个特定动作，如走路，可以利用上述特征提取方法获取特征。针对一个动作实例的特征，可以采用矢量量化的方法先对每帧信号强度进行量化，也可以直接利用混合高斯隐马尔可夫模型进行训练。在训练过程中，可以用传统的期望最大化算法来迭代生成动作模型。离线训练针对每个具体动作生成一个对应的隐马尔可夫模型。该模型可以用于在线的动作识别。
在线识别系统在进行动作检测和特征提取后，将提取的各帧信号强度矢量分别输入多个隐马尔可夫模型。每个隐马尔可夫模型对应一个动作。对每一个隐马尔可夫模型，系统计算其产生当前信号强度矢量序列的可能性。系统选择具有最大可能生成当前信号矢量序列的模型对应的动作作为识别结果。
虽然本发明已以较佳实施例公开如上，但实施例并不是用来限定本发明的。在不脱离本发明之精神和范围内，所做的任何等效变化或润饰，同样属于本发明之保护范围。因此本发明的保护范围应当以本申请的权利要求所界定的内容为标准。

Claims (10)

  1. 一种基于无线信号的动作检测和识别的方法，其特征在于：采用通用无线设备采集无线信号数据，对数据进行去噪处理后提取与运动速度相关联的特征对动作进行识别。
  2. 根据权利要求1所述的方法，其特征在于：使用一个或多个通用无线设备采集无线信号数据，通过多路无线信号数据的相关性对无线数据进行去噪处理，并从无线数据中提取出与人体运动速度相关的特征对运动进行检测和识别。
  3. 根据权利要求2所述的方法，其特征在于，它包括具体步骤为：
    数据采集步骤，通过对接收到的无线信号的强度进行测量；
    数据去噪步骤，对多路信号进行主成分分析(PCA,Principal component analysis)，获取噪声最低的数据成分；
    运动特征提取步骤，采用对无线信号数据进行时频分析获取无线信号在不同频率上的强度；
    模型训练步骤，系统对离线数据集的训练，通过采集不同动作对应的无线信号来建立训练数据集；
    动作识别步骤，动作识别方法采用隐马尔可夫模型(HMM，Hidden Markov Model)实现。
  4. 根据权利要求3所述的方法，其特征在于：所述通用无线设备是指支持WiFi，长期演进(LTE,Long Term Evolution)，蓝牙(Bluetooth)，Zigbee通信技术的无线设备。
  5. 根据权利要求3所述的方法，其特征在于：所述无线信号强度数据包括接收信号强度指示(RSSI，Received signal strength indication)或信道状态指示(CSI，Channel state indication)。
  6. 根据权利要求3所述的方法，其特征在于：
    数据去噪步骤中，所述多路信号是指：一、无线信号采用正交频分复用(OFDM，Orthogonal Frequency Division Multiplexing)调制时，多个子载波 上的信号强度能够单独测量；二、当无线发射器或无线接收器拥有多根天线时，每对发射/接收天线对上的信号强度能够单独测量；三、当有多个无线发射器或多个无线接收器时，每对发射/接收器上的信号强度能够单独测量；
    这些单独测量的无线数据均看成是独立的多路信号。
  7. 根据权利要求6所述的方法，其特征在于：
    针对多路信号采用联合去噪算法，依据人体运动在多路信号中的相关性，使用主成分分析的方法来从多路信号中提取出与运动相关的信息；
    所述使用主成分分析的具体方法为：在每一个子载波上无线接收器都能够测量一系列的信号强度，将不同子载波上的信号强度按照时间和子载波的顺序排列；随后对单个子载波上的信号强度序列进行预处理，预处理中将信号强度时间序列减去信号强度的长期平均值，以得到预处理后的信号强度；将预处理后的信号强度进行分段处理，获取测量数据矩阵H；测量矩阵H的每一行表示的是单个子载波上的信号强度，每一列上为同一时刻测量的不同子载波上的信号强度；测量矩阵H的行数n等于子载波的数量，测量矩阵的列数m等于时间序列的长度；对测量矩阵H进行相关操作可以获取其相关矩阵，相关矩阵的维度是n×n，对相关矩阵C进行奇异值分解(SVD Singular value decomposition)或特征值分解(eigendecomposition)，获取相关矩阵的特征值和特征向量；测量矩阵H与特征向量相乘获取信号的各个PCA成分。
  8. 根据权利要求7所述的方法，其特征在于，在数据去噪步骤和运动特征提取步骤之间还设有数据分段步骤：
    数据经过去噪处理后，需要对动作进行检测和分段，利用信号的方差信息来判断是否有动作；当有动作发生时，无线信号强度将有明显波动，通过其方差变大来进行判断，或者通过PCA的特征向量的平滑度来进行判断；
    当判断有动作发生后，对动作进行切分；在识别时依据分段后的动作信息来识别具体动作。
  9. 根据权利要求3所述的方法，其特征在于：
    运动特征提取步骤中，所述时频分析方法包括短时傅立叶变换(STFT，Short-time Fourier transformation)和小波变换(Wavelet transform)；
    短时傅立叶变换通过对信号加窗的方式，将信号分帧，计算每一帧信号的傅立叶变换值，获取信号在不同频率上的强度，每一帧信号将产生一个强度矢量，矢量中每一个数据表示信号在本帧中的某个频率上的强度；
    所述小波变换获取每一帧信号在各个频段上的信号强度矢量。
  10. 根据权利要求3所述的方法，其特征在于，所述动作识别步骤：
    首先，系统进行离线数据集的训练，通过采集不同动作对应的无线信号来建立训练数据集，在采集过程中，对信号进行标注和分段，标注过程是标明特定的信号是属于那一种动作，分段是人工将动作开始和结束点标出，形成训练数据集；
    系统在进行动作检测和特征提取后，将提取的各帧信号强度矢量分别输入多个隐马尔可夫模型，每个隐马尔可夫模型对应一个动作，对每一个隐马尔可夫模型，系统计算其产生当前信号强度矢量序列的可能性，系统选择具有最大可能生成当前信号矢量序列的模型对应的动作作为识别结果。
PCT/CN2016/076575 2015-06-10 2016-03-17 一种基于无线信号的动作检测和识别的方法 WO2016197648A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201510316143.8A CN104951757B (zh) 2015-06-10 2015-06-10 一种基于无线信号的动作检测和识别的方法
CN2015103161438 2015-06-10

Publications (1)

Publication Number Publication Date
WO2016197648A1 true WO2016197648A1 (zh) 2016-12-15

Family

ID=54166399

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/076575 WO2016197648A1 (zh) 2015-06-10 2016-03-17 一种基于无线信号的动作检测和识别的方法

Country Status (2)

Country Link
CN (1) CN104951757B (zh)
WO (1) WO2016197648A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108234043A (zh) * 2018-01-05 2018-06-29 中国矿业大学 一种煤矿井下关键区域人员闯入监测系统及其方法
EP3379850A1 (en) * 2017-03-21 2018-09-26 Nokia Technologies Oy Method and device for activity recognition
CN110414468A (zh) * 2019-08-05 2019-11-05 合肥工业大学 WiFi环境下基于手势信号的身份验证方法
CN112380903A (zh) * 2020-10-14 2021-02-19 东北电力大学 一种基于WiFi-CSI信号增强的人体活动识别方法
CN112990026A (zh) * 2021-03-19 2021-06-18 西北大学 基于对抗训练的无线信号感知模型构建、感知方法及系统
CN113259029A (zh) * 2021-05-04 2021-08-13 中国人民解放军32802部队 一种适用于无人机信号的实时自动检测识别方法
US11196492B2 (en) 2019-04-24 2021-12-07 Robert Bosch Gmbh Apparatus for person identification and motion direction estimation

Families Citing this family (36)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104951757B (zh) * 2015-06-10 2018-11-09 南京大学 一种基于无线信号的动作检测和识别的方法
CN105405260B (zh) * 2015-12-15 2018-06-26 刘向阳 一种基于无线信号的防盗系统及防盗方法
CN105761407B (zh) * 2016-01-06 2019-05-10 深圳大学 基于无线网络信号传输的室内探测火情及报警方法及系统
CN105933080B (zh) * 2016-01-20 2020-11-03 北京大学 一种跌倒检测方法和系统
CN105809110A (zh) * 2016-02-24 2016-07-27 南京大学 一种基于无线信号强度的行为识别系统与方法
CN106323330B (zh) * 2016-08-15 2019-01-11 中国科学技术大学苏州研究院 基于WiFi动作识别系统的非接触式计步方法
CN106411433B (zh) * 2016-09-08 2019-12-06 哈尔滨工程大学 一种基于wlan的细粒度室内被动入侵检测方法
CN107886103B (zh) * 2016-09-29 2023-12-08 日本电气株式会社 用于识别行为模式的方法、设备和系统
CN106936526A (zh) * 2017-03-30 2017-07-07 西北工业大学 一种基于信道状态信息的非接触式睡眠分期装置及方法
CN108809743B (zh) * 2017-04-28 2021-10-22 国基电子（上海）有限公司 物体侦测方法及网络装置
CN107331136A (zh) * 2017-05-11 2017-11-07 深圳市斑点猫信息技术有限公司 基于WiFi的室内人体活动检测方法和系统
TWI612828B (zh) * 2017-06-27 2018-01-21 樹德科技大學 居家空間跌倒偵測系統及其方法
CN107589725B (zh) * 2017-08-29 2020-03-24 深圳市盛路物联通讯技术有限公司 一种基于内置天线的智能家居设备控制方法及控制设备
CN107633227B (zh) * 2017-09-15 2020-04-28 华中科技大学 一种基于csi的细粒度手势识别方法和系统
CN107749143B (zh) * 2017-10-30 2023-09-19 安徽工业大学 一种基于WiFi信号的穿墙室内人员跌倒探测系统及方法
CN107862295B (zh) * 2017-11-21 2021-04-02 武汉大学 一种基于WiFi信道状态信息识别面部表情的方法
US10108903B1 (en) * 2017-12-08 2018-10-23 Cognitive Systems Corp. Motion detection based on machine learning of wireless signal properties
CN108553108B (zh) * 2018-03-05 2020-04-14 上海百芝龙网络科技有限公司 一种基于Wi-Fi中CSI信号的人体动作与呼吸的检测方法和系统
CN108901021B (zh) * 2018-05-31 2021-05-11 大连理工大学 一种基于无线网络信道状态信息的深度学习身份识别系统及方法
CN108805194B (zh) * 2018-06-04 2021-12-31 上海交通大学 一种基于wifi信道状态信息的手写识别方法及系统
CN109461295B (zh) * 2018-12-07 2021-06-11 连尚（新昌）网络科技有限公司 一种家居报警方法及设备
CN109766951A (zh) * 2019-01-18 2019-05-17 重庆邮电大学 一种基于时频统计特性的WiFi手势识别
CN110069134B (zh) * 2019-03-29 2020-11-27 北京大学 一种利用无线电射频信号还原手部空中移动轨迹的方法
CN111751814A (zh) 2019-03-29 2020-10-09 富士通株式会社 基于无线信号的运动状态检测装置、方法及系统
CN110176968B (zh) * 2019-05-20 2021-04-06 桂林理工大学 一种用于WiFi人体行为识别中的跳变现象纠正方法
CN110737201B (zh) * 2019-10-11 2020-10-09 珠海格力电器股份有限公司 一种监护方法、装置、存储介质及空调
CN110751115B (zh) * 2019-10-24 2021-01-01 北京金茂绿建科技有限公司 一种无接触式人体行为识别方法和系统
CN110852266A (zh) * 2019-11-11 2020-02-28 重庆邮电大学 一种基于无线信号的步态特征提取方法
CN111478950B (zh) * 2020-03-26 2023-04-18 微民保险代理有限公司 一种对象状态的推送方法和装置
CN111815906B (zh) * 2020-07-30 2022-03-11 苏州苗米智能技术有限公司 基于无线信号识别的摔倒监测方法和系统
CN112163540B (zh) * 2020-10-09 2024-01-19 上海第二工业大学 一种基于WiFi的姿态识别方法
CN112446426A (zh) * 2020-11-23 2021-03-05 中国科学技术大学 摔倒检测方法、装置、电子设备及存储介质
CN112596024B (zh) * 2020-12-04 2021-10-08 华中科技大学 一种基于环境背景无线射频信号的运动识别方法
TWI773038B (zh) * 2020-12-22 2022-08-01 中華電信股份有限公司 基於無線訊號強度之失能通報系統和失能通報方法
CN113238659A (zh) * 2021-06-29 2021-08-10 中国科学技术大学 基于wifi信号的实时行为识别方法与系统
CN115002710A (zh) * 2022-05-20 2022-09-02 海信集团控股股份有限公司 运动监测方法及电子设备

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2013130067A1 (en) * 2012-02-29 2013-09-06 Intel Corporation Detection of device motion and nearby object motion
US20140073346A1 (en) * 2012-09-07 2014-03-13 Lei Yang Proximity human motion detection using wireless signals
CN103971108A (zh) * 2014-05-28 2014-08-06 北京邮电大学 基于无线通信的人体姿态识别方法与装置
CN104504396A (zh) * 2014-12-18 2015-04-08 大连理工大学 利用自然环境无线信号的人体位置状态识别方法
CN104951757A (zh) * 2015-06-10 2015-09-30 南京大学 一种基于无线信号的动作检测和识别的方法

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101794528B (zh) * 2010-04-02 2012-03-14 北京大学软件与微电子学院无锡产学研合作教育基地 一种手语语音双向翻译系统
US8373658B2 (en) * 2010-05-24 2013-02-12 Cywee Group Limited Motion sensing system

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2013130067A1 (en) * 2012-02-29 2013-09-06 Intel Corporation Detection of device motion and nearby object motion
US20140073346A1 (en) * 2012-09-07 2014-03-13 Lei Yang Proximity human motion detection using wireless signals
CN103971108A (zh) * 2014-05-28 2014-08-06 北京邮电大学 基于无线通信的人体姿态识别方法与装置
CN104504396A (zh) * 2014-12-18 2015-04-08 大连理工大学 利用自然环境无线信号的人体位置状态识别方法
CN104951757A (zh) * 2015-06-10 2015-09-30 南京大学 一种基于无线信号的动作检测和识别的方法

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP3379850A1 (en) * 2017-03-21 2018-09-26 Nokia Technologies Oy Method and device for activity recognition
CN108234043A (zh) * 2018-01-05 2018-06-29 中国矿业大学 一种煤矿井下关键区域人员闯入监测系统及其方法
US11196492B2 (en) 2019-04-24 2021-12-07 Robert Bosch Gmbh Apparatus for person identification and motion direction estimation
CN110414468A (zh) * 2019-08-05 2019-11-05 合肥工业大学 WiFi环境下基于手势信号的身份验证方法
CN112380903A (zh) * 2020-10-14 2021-02-19 东北电力大学 一种基于WiFi-CSI信号增强的人体活动识别方法
CN112380903B (zh) * 2020-10-14 2024-02-02 东北电力大学 一种基于WiFi-CSI信号增强的人体活动识别方法
CN112990026A (zh) * 2021-03-19 2021-06-18 西北大学 基于对抗训练的无线信号感知模型构建、感知方法及系统
CN112990026B (zh) * 2021-03-19 2024-01-19 西北大学 基于对抗训练的无线信号感知模型构建、感知方法及系统
CN113259029A (zh) * 2021-05-04 2021-08-13 中国人民解放军32802部队 一种适用于无人机信号的实时自动检测识别方法
CN113259029B (zh) * 2021-05-04 2022-03-22 中国人民解放军32802部队 一种适用于无人机信号的实时自动检测识别方法

Also Published As

Publication number Publication date
CN104951757B (zh) 2018-11-09
CN104951757A (zh) 2015-09-30

Similar Documents

Publication Publication Date Title
WO2016197648A1 (zh) 一种基于无线信号的动作检测和识别的方法
Wang et al. Device-free human activity recognition using commercial WiFi devices
Arshad et al. Wi-chase: A WiFi based human activity recognition system for sensorless environments
Zhou et al. Device-free presence detection and localization with SVM and CSI fingerprinting
Zhang et al. Wifi-id: Human identification using wifi signal
Qian et al. PADS: Passive detection of moving targets with dynamic speed using PHY layer information
WO2018133264A1 (zh) 一种人体室内定位自动检测方法及系统
Huang et al. Indoor device-free activity recognition based on radio signal
KR100948412B1 (ko) 수신 신호 강도를 이용한 위치추정 방법 및 시스템
Miyazaki et al. Initial attempt on outdoor human detection using IEEE 802.11 ac WLAN signal
CN104502894A (zh) 基于物理层信息的运动物体被动检测方法
Liu et al. A research on CSI-based human motion detection in complex scenarios
Cao et al. Wi-Wri: Fine-grained writing recognition using Wi-Fi signals
Shi et al. Human activity recognition using deep learning networks with enhanced channel state information
Chowdhury et al. WiHACS: Leveraging WiFi for human activity classification using OFDM subcarriers' correlation
CN112330924B (zh) 一种室内环境下基于信道状态信息的跌倒事件检测方法
Zhu et al. NotiFi: A ubiquitous WiFi-based abnormal activity detection system
Cheng et al. Device-free human activity recognition based on GMM-HMM using channel state information
Hao et al. CSI‐HC: A WiFi‐Based Indoor Complex Human Motion Recognition Method
CN109947238A (zh) 一种基于wifi的非合作式手势识别的方法
Muaaz et al. WiHAR: From Wi-Fi channel state information to unobtrusive human activity recognition
Liu et al. Wi-CR: human action counting and recognition with Wi-Fi signals
Man et al. PWiG: A phase-based wireless gesture recognition system
Shen et al. WiRIM: Resolution improving mechanism for human sensing with commodity Wi-Fi
Yousefi et al. A survey of human activity recognition using wifi CSI

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 16806546

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 16806546

Country of ref document: EP

Kind code of ref document: A1