CN111542012B - Human body tumbling detection method based on SE-CNN - Google Patents

Human body tumbling detection method based on SE-CNN Download PDF

Info

Publication number
CN111542012B
CN111542012B CN202010347163.2A CN202010347163A CN111542012B CN 111542012 B CN111542012 B CN 111542012B CN 202010347163 A CN202010347163 A CN 202010347163A CN 111542012 B CN111542012 B CN 111542012B
Authority
CN
China
Prior art keywords
cnn
wavelet
domain
human body
radio frequency
Prior art date
Legal status (The legal status 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 status listed.)
Expired - Fee Related
Application number
CN202010347163.2A
Other languages
Chinese (zh)
Other versions
CN111542012A (en
Inventor
杨志勇
谷长春
金磊
马腾飞
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanchang Hangkong University
Original Assignee
Nanchang Hangkong University
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 Nanchang Hangkong University filed Critical Nanchang Hangkong University
Priority to CN202010347163.2A priority Critical patent/CN111542012B/en
Publication of CN111542012A publication Critical patent/CN111542012A/en
Application granted granted Critical
Publication of CN111542012B publication Critical patent/CN111542012B/en
Expired - Fee Related legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04WWIRELESS COMMUNICATION NETWORKS
    • H04W4/00Services specially adapted for wireless communication networks; Facilities therefor
    • H04W4/30Services specially adapted for particular environments, situations or purposes
    • H04W4/38Services specially adapted for particular environments, situations or purposes for collecting sensor information
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04BTRANSMISSION
    • H04B17/00Monitoring; Testing
    • H04B17/30Monitoring; Testing of propagation channels
    • H04B17/309Measuring or estimating channel quality parameters
    • H04B17/318Received signal strength
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/08Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Quality & Reliability (AREA)
  • Physics & Mathematics (AREA)
  • Electromagnetism (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)

Abstract

The invention discloses a human body tumble detection method based on SE-CNN, which comprises the following steps of (1) establishing a data acquisition environment: a wireless radio frequency tomography network node is used as a communication basis of the network; constructing a complete wireless sensor network communication system by using network nodes; (2) collecting and processing data: the processing method of the wireless sensor network data is 'transmitting-receiving-storing'; extracting effective links, denoising, wavelet transformation, time domain characteristics and wavelet domain characteristics according to the collected radio frequency signal intensity values, and training an SE-CNN model by using the extracted multi-domain characteristics; (3) the real-time human body tumbling detection method comprises the following steps: and screening the multi-domain features obtained by the XGboost model in a permutation and combination mode on the time domain feature components and the wavelet domain feature components to obtain joint feature components with strong robustness, and establishing a multi-domain feature perception fingerprint library of the falling action. And the SE-CNN is trained by using the extracted multi-domain characteristics, so that the detection of human body tumble can be realized. The invention has simple structure, is convenient and feasible in implementation and is suitable for most scenes.

Description

Human body tumbling detection method based on SE-CNN
Technical Field
The invention relates to the technical field of wireless sensor networks, in particular to a human body tumbling detection method based on SE-CNN.
Background
The ever-aging population worldwide presents a significant challenge to the system of care and prevention of household accidents, particularly for elderly people living alone. According to the latest statistics, the aged people above 60 years old reach 2.4 hundred million and account for 17.3 percent of the total population by 2017 according to the data of China aged offices. This ratio is also rising considering people over age 70. The direct consequence of a fall may be muscle or ligament injury, fracture and head trauma, and consequent brain damage, among others. Serious injury constitutes a significant risk to morbidity and mortality following a fall.
With the advancement of technology, wireless signals have substantially covered every corner of our lives. With the rapid development of artificial intelligence technology in recent years, the role of wireless signals is not limited to the traditional communication field, and is endowed with more possibilities, thus receiving wide attention and driving the development of a series of emerging technologies. In the area covered by the wireless network, human activities can affect wireless signal links, such as physical actions of absorption, reflection, refraction and the like of wireless signals. Different human body actions have different degrees of shielding the wireless link, and therefore different influences can be caused certainly. The human body falling detection is carried out by utilizing a radio frequency wireless sensor network formed by radio frequency signals, appropriate sensor nodes are arranged in a monitoring area, the sensor nodes are self-organized into the wireless sensor network, the state characteristic component of a monitoring target is extracted by adopting a corresponding technology, and the falling state of the human body target is judged by combining a deep learning method. In the monitoring process, a human target does not need to carry equipment, and video or photographing monitoring is not needed, so that the privacy of a user can be protected to a great extent in a bedroom or a toilet where the falling probability is large and the user is private, and the acceptance of the user to a wireless signal technology is increased. The wireless radio frequency signal technology is used for detecting the falling of the human body, has the characteristics of easiness in arrangement, low cost, privacy protection and the like, has good research value and commercial value, and has wide application scenes in the arrangement of accelerating smart cities, intelligent medical treatment and smart homes in the future.
Disclosure of Invention
The invention aims to solve the problems that: the human body tumble detection method based on the SE-CNN is simple in structure, convenient and feasible in implementation and suitable for most scenes.
The technical scheme provided by the invention for solving the problems is as follows: a human body tumble detection method based on SE-CNN comprises the following steps,
(1) and building a data acquisition environment: a wireless radio frequency tomography network node is used as a communication basis of the network; constructing a complete wireless sensor network communication system by using network nodes;
(2) collecting and processing data: the processing method of the wireless sensor network data is 'transmitting-receiving-storing'; extracting effective links, denoising, wavelet transformation, time domain characteristics and wavelet domain characteristics according to the collected radio frequency signal intensity values, and training an SE-CNN model by using the extracted multi-domain characteristics;
(3) the real-time human body tumbling detection method comprises the following steps: and screening the multi-domain features obtained by the XGboost model in a permutation and combination mode on the time domain feature components and the wavelet domain feature components to obtain combined feature components with strong robustness, and establishing a multi-domain feature perception fingerprint library of the falling action. And the SE-CNN is trained by using the extracted multi-domain characteristics, so that the detection of human body tumble can be realized.
Preferably, the wireless network system mainly comprises 6 PVC pipes, 36 sensor nodes, 1 aggregation node and a PC.
Preferably, the "transmission-reception-storage" mode is as follows:
(1) the power supply is switched on, and the sensor nodes periodically transmit the radio frequency signal intensity of the link formed by the nodes in a polling mode;
(2) and the periodic radio frequency signal sent by the sensor node is received by a sink node connected to the PC and is transmitted to the PC through the serial assistant to be stored as a txt text.
1. Preferably, the SE-CNN network classification model algorithm is:
(1) and data preprocessing: deleting incomplete periods before and after each data text to obtain a plurality of complete periods of data, converting hexadecimal into decimal, and subtracting an offset to obtain a real RSS value;
(2) and extracting a characteristic value: deleting an invalid link, extracting time domain characteristics from the RSS measured value of the valid link, then using wavelet transformation to process the RSS measured value of the valid link to extract more robust wavelet characteristics, and then utilizing an XGboost model to respectively compare the characteristics of different combinations to select the time domain wavelet combination characteristics with the most distinguishing capability;
(3) modeling and parameter solving: the time domain wavelet joint characteristics and the source RSS measurement value are obtained and used as input data of the SE-CNN method, the characteristics of the wireless radio frequency signal input matrix are extracted by the SE-CNN, and the mutual dependency relationship of characteristic channels of the wireless radio frequency signal input matrix is explicitly modeled.
Preferably, real-time data is collected and time domain characteristics of the RSS measurement values of the effective link are extracted by continuously monitoring in real time, and then more robust wavelet characteristics are extracted by processing the RSS measurement values of the effective link by using wavelet transformation. And taking the acquired time domain wavelet joint characteristics and the source RSS measured value as input data of the SE-CNN model. And establishing a fingerprint database so as to detect the condition of the human body in real time.
Compared with the prior art, the invention has the advantages that: the work applies radio frequency signals to people flow rate monitoring, uses a novel wireless radio frequency tomography network node as a communication node, utilizes the characteristic that the radio frequency signals can be influenced by a human body and combines a machine learning method to monitor people flow rate, is simple in structure, convenient and feasible in realization and suitable for most scenes.
Drawings
The accompanying drawings, which are included to provide a further understanding of the invention and are incorporated in and constitute a part of this specification, illustrate embodiments of the invention and together with the description serve to explain the invention and not to limit the invention.
FIG. 1 is a diagram of the monitoring system configuration of the present invention;
FIG. 2 is a schematic diagram of the wavelet transform of the present invention;
FIG. 3 is a schematic diagram of the SE-CNN-based classification performance comparison of the present invention;
Detailed Description
The following detailed description of the embodiments of the present invention will be provided with reference to the accompanying drawings and examples, so that how to implement the embodiments of the present invention by using technical means to solve the technical problems and achieve the technical effects can be fully understood and implemented.
A human body tumble detection method based on SE-CNN comprises the following steps,
(1) and building a data acquisition environment: a wireless radio frequency tomography network node is used as a communication basis of the network; constructing a complete wireless sensor network communication system by using network nodes; the node is a low-power-consumption wireless radio frequency network node platform suitable for deployment in smaller scenes such as intrusion detection, people flow statistics and falling monitoring. The low-power-consumption wireless radio frequency network node platform takes a system-on-chip (SoC) chip CC2530 as a core, and 2 programmable LED lamps and 1 key are expanded; the onboard voltage stabilizing module is compatible with the most common 3.3-5V power supply; and the JTAG interface of the programming program is designed as a non-standard interface with only 5 pins (which can supply power to the nodes) so as to reduce the size of the nodes as much as possible, namely, only 24 x 31 mm. Therefore, the low-power-consumption wireless radio frequency network node platform has the characteristics of low cost, small size, light weight, low power consumption, high reliability and the like, and is very suitable for the installation and deployment of indoor intelligent home and intelligent sensing-oriented wireless radio frequency tomography networks.
(2) Collecting and processing data: the processing method of the wireless sensor network data is 'transmitting-receiving-storing'; extracting effective links, denoising, wavelet transformation and time domain characteristics and wavelet domain characteristics according to the collected radio frequency signal intensity values;
(3) the real-time human body tumbling detection method comprises the following steps: and screening the multi-domain features obtained by the XGboost model in a permutation and combination mode on the time domain feature components and the wavelet domain feature components to obtain joint feature components with strong robustness, and establishing a multi-domain feature perception fingerprint library of the falling action. And the extracted multi-domain characteristics are used for training the SE-CNN model, so that the detection of human body tumble can be realized.
The wireless network system mainly comprises 6 PVC pipes, 36 sensor nodes, 1 aggregation node and a PC.
The transmission-receiving-storing mode is as follows:
(1) the power supply is switched on, and the sensor nodes periodically transmit the radio frequency signal intensity of the link formed by the nodes in a polling mode;
(2) and the periodic radio frequency signal sent by the sensor node is received by a sink node connected to the PC and is transmitted to the PC through the serial assistant to be stored as a txt text.
2. The SE-CNN network classification model algorithm is as follows:
(1) and data preprocessing: deleting each data text to obtain incomplete periods, enabling the obtained data to be a plurality of complete periods, collecting enough RSS sample data and filling missing values;
(2) and extracting a characteristic value: deleting an invalid link, extracting time domain characteristics from the RSS measured value of the valid link, then using wavelet transformation to process the RSS measured value of the valid link to extract more robust wavelet characteristics, and then utilizing an XGboost model to respectively compare the characteristics of different combinations to select the time domain wavelet combined characteristics with the most distinguishing capability. (ii) a
(3) And establishing a model: the acquired time domain wavelet domain joint characteristics and the source RSS measurement value are used as input data of an SE-CNN model, the characteristics of a wireless radio frequency signal input matrix are extracted by the SE-CNN model, and the mutual dependency relationship of characteristic channels of the wireless radio frequency signal input matrix is explicitly modeled.
The algorithm is realized according to the following principle: the resonance frequency of the radio frequency signal is consistent with that of water at 2.4GHz, and the human body can generate great influence on the radio frequency signal due to the fact that most of the human body consists of water, and the algorithm is provided.
Real-time data is collected and the time domain characteristics of the RSS measured values of the effective links are extracted through continuous real-time monitoring, and then the wavelet transform is used for processing the RSS measured values of the effective links to extract more robust wavelet characteristics. And taking the acquired time domain wavelet domain joint characteristics and the source RSS measured value as input data of the SE-CNN model. And establishing a fingerprint database so as to detect the condition of the human body in real time. The invention has the beneficial effects that:
the foregoing is merely illustrative of the preferred embodiments of the present invention and is not to be construed as limiting the claims. The present invention is not limited to the above embodiments, and the specific structure thereof is allowed to vary. All changes which come within the scope of the invention as defined by the independent claims are intended to be embraced therein.

Claims (3)

1. A human body tumbling detection method based on SE-CNN is characterized in that:
(1) and building a data acquisition environment: a wireless radio frequency tomography network node is used as a communication basis of the network; constructing a complete wireless sensor network communication system by using the network node;
(2) collecting and processing data: the processing method of the wireless sensor network data is 'transmitting-receiving-storing'; extracting effective links, denoising, wavelet transformation, time domain characteristics and wavelet domain characteristics according to the collected radio frequency signal intensity values, and training an SE-CNN model by using the extracted multi-domain characteristics;
(3) the transmission-receiving-storage mode is as follows:
1) the power supply is switched on, and the sensor nodes periodically transmit the radio frequency signal intensity of a link formed by each node in a polling mode;
2) the periodic radio frequency signal sent by the sensor node is received by a sink node connected to the PC and is transmitted to the PC through a serial assistant to be stored as a txt text;
(4) the real-time human body tumbling detection method comprises the following steps: screening multi-domain features obtained by arranging and combining the time domain feature components and the wavelet domain feature components by using an XGboost model to obtain joint feature components with strong robustness, and establishing a multi-domain feature perception fingerprint library of the falling action; the SE-CNN is trained by using the extracted multi-domain characteristics, so that the detection of human body tumble can be realized;
(5) the method is characterized in that: the SE-CNN model algorithm is as follows:
1) and data preprocessing: deleting incomplete periods before and after each data text to obtain a plurality of complete periods of data, converting hexadecimal into decimal, and subtracting an offset to obtain a real RSS value;
2) and extracting a characteristic value: deleting an invalid link, extracting time domain characteristics from the RSS measured value of the valid link, then using wavelet transformation to process the RSS measured value of the valid link to extract more robust wavelet characteristics, and then utilizing an XGboost model to respectively compare the characteristics of different combinations to select the time domain wavelet combination characteristics with the most distinguishing capability;
3) and establishing a model: the time domain wavelet joint characteristics and the source RSS measurement value are obtained and used as input data of the SE-CNN method, the characteristics of the wireless radio frequency signal input matrix are extracted by the SE-CNN, and the mutual dependency relationship of characteristic channels of the wireless radio frequency signal input matrix is explicitly modeled.
2. The SE-CNN-based human fall detection method of claim 1, wherein: the wireless sensor network communication system mainly comprises 6 PVC pipes, 36 sensor nodes, 1 aggregation node and a PC.
3. The SE-CNN-based human body tumbling detection method as claimed in claim 1, wherein the human body tumbling detection is realized by continuously performing real-time monitoring, collecting real-time data, extracting more robust wavelet features by processing effective link RSS measurement values through wavelet transformation, obtaining time domain wavelet domain joint features with strong robustness through screening by using an XGboost model, automatically obtaining the importance degree of feature channels through a deep learning network, and improving useful features according to the importance degree of the feature channels.
CN202010347163.2A 2020-04-28 2020-04-28 Human body tumbling detection method based on SE-CNN Expired - Fee Related CN111542012B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010347163.2A CN111542012B (en) 2020-04-28 2020-04-28 Human body tumbling detection method based on SE-CNN

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010347163.2A CN111542012B (en) 2020-04-28 2020-04-28 Human body tumbling detection method based on SE-CNN

Publications (2)

Publication Number Publication Date
CN111542012A CN111542012A (en) 2020-08-14
CN111542012B true CN111542012B (en) 2022-05-03

Family

ID=71980230

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010347163.2A Expired - Fee Related CN111542012B (en) 2020-04-28 2020-04-28 Human body tumbling detection method based on SE-CNN

Country Status (1)

Country Link
CN (1) CN111542012B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112545496A (en) * 2020-11-09 2021-03-26 金茂智慧科技(广州)有限公司 Wireless technology identification method for detecting human body tumble

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104080053A (en) * 2014-07-16 2014-10-01 上海诺诚电气有限公司 Human falling-down detecting and positioning system and method based on Zigbee technology
CN106776580A (en) * 2017-01-20 2017-05-31 中山大学 The theme line recognition methods of the deep neural network CNN and RNN of mixing
CN109214285A (en) * 2018-08-01 2019-01-15 浙江深眸科技有限公司 Detection method is fallen down based on depth convolutional neural networks and shot and long term memory network
CN109600758A (en) * 2018-11-15 2019-04-09 南昌航空大学 A kind of stream of people's quantity monitoring method based on RSS
CN110674875A (en) * 2019-09-25 2020-01-10 电子科技大学 Pedestrian motion mode identification method based on deep hybrid model
CN111028477A (en) * 2019-12-06 2020-04-17 哈尔滨理工大学 Intelligent tumble detection device and method based on convolutional neural network

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104080053A (en) * 2014-07-16 2014-10-01 上海诺诚电气有限公司 Human falling-down detecting and positioning system and method based on Zigbee technology
CN106776580A (en) * 2017-01-20 2017-05-31 中山大学 The theme line recognition methods of the deep neural network CNN and RNN of mixing
CN109214285A (en) * 2018-08-01 2019-01-15 浙江深眸科技有限公司 Detection method is fallen down based on depth convolutional neural networks and shot and long term memory network
CN109600758A (en) * 2018-11-15 2019-04-09 南昌航空大学 A kind of stream of people's quantity monitoring method based on RSS
CN110674875A (en) * 2019-09-25 2020-01-10 电子科技大学 Pedestrian motion mode identification method based on deep hybrid model
CN111028477A (en) * 2019-12-06 2020-04-17 哈尔滨理工大学 Intelligent tumble detection device and method based on convolutional neural network

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
Wavelet Package Analysis based Fall Detection and Diagnosis;Meichuan Huang等;《Proceedings of the 37th Chinese Control Conference》;20180727;全文 *
WiFall: Device-Free Fall Detection by Wireless Networks;Yuxi Wang等;《IEEE TRANSACTIONS ON MOBILE COMPUTING》;20170228;全文 *
基于时序分析的人体摔倒预测方法;佟丽娜等;《模式识别与人工智能》;20120415(第02期);全文 *
采用卷积神经网络的老年人跌倒检测系统设计;吕艳等;《浙江大学学报(工学版)》;20190513(第06期);全文 *

Also Published As

Publication number Publication date
CN111542012A (en) 2020-08-14

Similar Documents

Publication Publication Date Title
Akmandor et al. Smart, secure, yet energy-efficient, Internet-of-Things sensors
CN204580691U (en) A kind of intelligent quilt
CN103989295B (en) Electronic wrist strap system and the method for detecting abnormality of personnel's management and control are corrected in Community-oriented
CN103093234B (en) Based on the personal identification method of ground reaction force during walking
CN111542012B (en) Human body tumbling detection method based on SE-CNN
CN109089077A (en) A kind of method remotely monitored and monitoring client
Suryadevara et al. Wireless sensors network based safe home to care elderly people: behaviour detection
CN109600758B (en) RSS-based people flow monitoring method
CN203250230U (en) Multifunctional wireless device of greenhouse environment factor mobile phone remote real-time monitoring
Sabo et al. The event-driven power efficient wireless sensor nodes for monitoring of insects and health of plants
Sabo et al. An event driven wireless sensors network for monitoring of plants health and larva activities
CN203645709U (en) Kindergarten intelligent monitoring system based on piezoelectric effect
Rahman et al. Wireless sensor node design
CN106448051A (en) Wearable physiological sensation equipment applicable to high-altitude operation protection
CN107172385A (en) A kind of internet of things intelligent household monitoring system
CN208010403U (en) Coal mine underground environmental safety monitors and personnel location system
CN214851933U (en) Infrared human body induction wireless controller
CN206096849U (en) Intelligence body -building venue data acquisition system
CN211718711U (en) Empty nest old man safety control system at home based on thing networking
CN108389355A (en) A kind of low energy consumption Smoke Detection warning device based on NB-IoT technologies
CN105832298A (en) Wireless intelligent dynamic medical monitoring and analyzing system
Ye Intelligent Inspection System of Power Equipment Based on Photoelectric Sensor/AR Technology
CN208689406U (en) A kind of farmland nurse equipment and system
CN208622224U (en) A kind of swift Smoke Detection warning device based on NB-IoT technology
Liu et al. Intelligent home design for elderly health monitoring and rating based wireless passive sensor network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant
CF01 Termination of patent right due to non-payment of annual fee

Granted publication date: 20220503

CF01 Termination of patent right due to non-payment of annual fee