CN110245588A - A kind of fine granularity estimation method of human posture based on radio frequency signal - Google Patents
A kind of fine granularity estimation method of human posture based on radio frequency signal Download PDFInfo
- Publication number
- CN110245588A CN110245588A CN201910457667.7A CN201910457667A CN110245588A CN 110245588 A CN110245588 A CN 110245588A CN 201910457667 A CN201910457667 A CN 201910457667A CN 110245588 A CN110245588 A CN 110245588A
- Authority
- CN
- China
- Prior art keywords
- posture
- network
- wifi
- human
- human body
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
-
- G—PHYSICS
- G08—SIGNALLING
- G08C—TRANSMISSION SYSTEMS FOR MEASURED VALUES, CONTROL OR SIMILAR SIGNALS
- G08C17/00—Arrangements for transmitting signals characterised by the use of a wireless electrical link
- G08C17/02—Arrangements for transmitting signals characterised by the use of a wireless electrical link using a radio link
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Biology (AREA)
- Computer Networks & Wireless Communication (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
A kind of fine granularity estimation method of human posture based on radio frequency signal disclosed by the invention, wifi are established using the wireless transmitting terminals of three antennas, when user walks or makes certain movement in the wifi area of coverage, specific influence can be generated on wifi channel, wireless receiving end receives wifi signal and calculates the CSI value of human body attitude, personage's video is captured using camera synchronous with the timestamp of WiFi antenna alignment, according to CSI value and video training deep learning network, utilize the network-evaluated human body attitude of trained deep learning, single Attitude estimation is carried out to realize using wifi equipment, convenience can be met while realizing higher recognition accuracy, ease for use and safety, and it does not need user and carries any special installation, the privacy life of user is not will record, With convenient easily deployment, highly-safe feature.
Description
Technical field
The present invention relates to human body attitude cognition technology fields, specially the fine granularity human body attitude based on radio frequency signal
Estimation method.
Background technique
Now with the development and the improvement of people's living standards of science and technology, smart home theory and virtual reality skill
Art is rapidly developed.In the crowd of the computer vision fields such as video monitoring, human-computer interaction, motion analysis and virtual reality
In more applications, human body is all main object processed, and the identification to human action and behavior is all one essential
Link, have urgent application demand, have become a very popular research direction in computer vision field.Human body
Attitude estimation be from the interconnection obtained in individual RGB image or video between human body bone point position and bone point, it is final defeated
The process of the whole or local limbs relevant parameter (relative positional relationship of each artis) of human body out, such as human body contour outline, head
The position in portion and direction, the position of human synovial and site categories etc..Human body attitude estimation intelligent monitoring, human-computer interaction and
The fields such as gesture identification have broad application prospects.
In order to carry out that it is big to be currently mainly used three as human body segmentation and this kind of fine-grained human perceptions of pose estimation
Class sensor: camera, radar and laser radar.These sensors can directly be caught in 2D image, depth map or 3D point cloud
Obtain the human body with high spatial resolution.For example, the video camera of 300*300 pixel, depth resolution is about 2 centimetres of radar,
Or 32 wave beam laser radar.
Based on the method that camera carries out human body segmentation and Attitude estimation, there can be high-altitude by Direct Acquisition in 2D image
Between resolution ratio human body.But also receive many limitations, for example, clothes, background, illuminate and block etc. technological challenges and
The limitation of the society such as privacy concern.
Based on the method that radar sensor carries out human body segmentation and Attitude estimation, Direct Acquisition can have in depth map
The human body of high spatial resolution.However, it is desirable to dedicated hardware, for example, frequency of use modulates continuous wave (FMCW) technology, broadband
Wide (1.78GHz) is assembled and synchronous 16+4T shape aerial array meticulously.
It, can be in 3D point cloud based on the method that laser radar sensor high-definition carries out human body segmentation and Attitude estimation
Middle Direct Acquisition has the human body of high spatial resolution.But laser radar is very expensive and power consumption, therefore, it is difficult to daily
Large-scale use in life and home environment.
Summary of the invention
Aiming at the problems existing in the prior art, the present invention provides a kind of fine granularity human body appearance based on radio frequency signal
State estimation method identifies human body appearance by analyzing the different changing patteries of channel state information using common commercial WiFi
State variation, carries out fine-grained human perception, can realize higher recognition accuracy, while meeting convenience and safety
's.
The present invention is to be achieved through the following technical solutions:
A kind of fine granularity estimation method of human posture based on radio frequency signal, comprising the following steps:
S1, using with more antennas wireless transmitting terminals and wireless receiving end establish the field wifi of waveform stabilization, simultaneously
In wireless transmitting terminals, one synchronization camera being aligned with wireless transmitting terminals timestamp is set；
S2, performance objective execute human body attitude in wifi, and wireless receiving end is received across performance objective, and is being held
The wifi signal for reflecting and reflecting in row target or surrounding objects, is calculated people according to the received wifi signal in wireless receiving end
The channel state information CSI value of body posture, meanwhile, the synchronous camera of timestamp obtains the video pictures of human body attitude, completes data
Collection；
S3, training dataset D is established according to the picture of channel state information CSI value and all video frames；
D={ (It,Ct),t∈[1,n]}；
Wherein, ItAnd CtThe respectively video frame of synchronization and CSI value, t indicate sampling instant, and n is data set size；
S4, building deep learning network；
S5, deep learning network is trained according to training dataset D, obtains trained deep learning network, root
Human body attitude estimation is carried out according to trained depth network.
Preferably, the method that channel state information CSI value is obtained in step S2 is specific as follows:
CSI=m × a × b × c
Wherein, m is the quantity of the WiFi packet received, and a is OFDM carrier number, and b and c are wireless transmitting terminals respectively and wireless
The antenna amount of receiving end.
Preferably, deep learning network described in step S3 includes teacher network and student network, and student network includes coding
Module, characteristic extracting module and decoder module；
Coding module, for by CSI value CtSize up-sample to RGB standard size；
Characteristic extracting module, for the C after up-samplingtFeature is extracted, and exports the CSI value F after extracting featuret；
Decoder module, for according to the CSI value F after extraction featuret, export the corresponding prediction matrix of posture adjacency matrix
pPAMt；
Teacher's network, for according to video frame ItExport posture adjacency matrix PAMt, to prediction matrix pPAMtIt exercises supervision
Optimization.
Preferably, teacher's network exports posture adjacency matrix PAMtMethod it is as follows:
Firstly, obtaining the side of performance objective in video frame using each frame picture in AlphaPose method processing video
Boundary's frame, then bounding box is returned into device by posture and is handled, the posture of performance objective in bounding box is obtained, and generate n
Three element predictions, n are the quantity of the artis to be estimated；
Then, by the posture of human body and three element predictive conversion posture adjacency matrix PAMt。
Preferably, the student network exports posture adjacency matrix pPAMtMethod it is as follows:
The coding module is CSI-Net network, using bilinear interpolation operation by Ct∈R150×3×3It is upsampled to
Ct∈R150×144×144；
Characteristic extracting module is 4 basic ResNet blocks in ResNets network, Ct∈R150×144×144By ResNets
The study of network exports Ft∈R300×18×18；
Decoder module is FCN network, inputs the CSI value F after feature extractiont, learn personage by two convolutional layers
Posture feature exports the corresponding prediction matrix pPAM of posture adjacency matrixt。
Preferably, using loss function L to prediction matrix pPAMtIt optimizes, specific as follows:
Wherein, pPAMxAnd PAMxIt is the prediction matrix of human joint points coordinate and posture adjacency matrix in x-axis respectively,
pPAMyAnd PAMyIt is the prediction matrix of human joint points coordinate and posture adjacency matrix in y-axis respectively.
Compared with prior art, the invention has the following beneficial technical effects:
Fine granularity estimation method of human posture provided by the invention based on radio frequency signal, using wireless transmitting terminals and
Wireless receiving end establishes wifi, when user walks or make certain movement in the wifi area of coverage, can produce to wifi channel
Raw specific to influence, wireless receiving end receives wifi signal and calculates the CSI value of human body attitude, at the same obtain user walking
Or action video, training dataset D is established according to the data of acquisition, deep learning network is instructed according to training dataset
Practice, obtain trained deep learning network, human body attitude estimation is carried out according to trained depth network, realizes and utilizes wifi
Equipment carries out single Attitude estimation, and the present invention can meet convenience, ease for use while realizing higher recognition accuracy
It with safety, and does not need user and carries any special installation, there is convenient easily deployment, highly-safe feature.
Further, deep learning network selects to use AlphaPose as teacher's network and WiSPPN as student's net
Network can carry out the general of damage human body attitude estimation caused by recurrence body joint point coordinate to avoid AlphaPose method is used only
Change capability problems, also avoids doing Signal averaging caused by human perception using only wifi equipment, obtained spatial information is more
The problem of coarseness.So that Attitude estimation can be realized higher recognition accuracy and fine granularity.
Detailed description of the invention
Fig. 1 is overall framework figure of the invention；
Fig. 2 is experiment scene deployment schematic diagram；
Fig. 3 is camera deployment schematic diagram；
Fig. 4 is camera schematic diagram synchronous with WiFi signal；
Fig. 5 is building posture adjacency matrix (PAM) process schematic；
Fig. 6 is a kind of block schematic illustration for new depth network that this work proposes；
Fig. 7 is the process schematic that feature extraction is completed using ResNets network；
Fig. 8 is the parameter schematic diagram of characteristic extraction procedure；
Fig. 9 is the process schematic being decoded using FCN network to CSI；
Figure 10 is experimental result picture.
Specific embodiment
Present invention will be described in further detail below with reference to the accompanying drawings, described to be explanation of the invention rather than limit
It is fixed.
Refering to fig. 1, a kind of fine granularity estimation method of human posture based on radio frequency signal, the specific steps are as follows:
Step 1, referring to fig. 2, wireless transmitting terminals use wifi transmitter and wireless transmitting terminals with three antennas to use
Wifi receiver with three antennas establishes the field wifi of waveform stabilization.
Step 2, referring to Fig. 3, a camera synchronous with the timestamp of wifi antenna alignment is disposed beside wireless transmitting terminals,
For capturing the video pictures of each frame of personage, Fig. 4 show camera video frame picture and WiFi signal time synchronization.
Human body appearance is executed in the field wifi that step 3, a people station are established between wireless transmitting terminals and wireless transmitting terminals
State, each pulse signal from transmitting terminal antenna broadcast to receiving end antenna, wifi signal penetrate human body or human body, furniture or/
It is reflected or/and is reflected with wall surface, wifi receiver is for receiving all wifi signals, while camera is used to capture people
The real-time pictures of body posture.
Step 4, when being pierced, reflect and when the signal that reflects reaches receiving antenna, refracted signal and reflection signal it is folded
Individual signals sample is added as, the channel state information CSI value of the human body attitude of each different people is calculated according to individual signals sample,
The CSI value of parsing is the tensor having a size of m × 30 × 3 × 3；
Wherein, m is the quantity of the WiFi packet received, and 30 be OFDM carrier number, most latter two 3 respectively represent sender and
The antenna amount of recipient completes data collection.
Step 5 establishes training dataset according to CSI value and video as D={ (It,Ct),t∈[1,n]}；
Wherein, ItAnd CtIt is a pair of of synchronized video frames and CSI value respectively, t indicates sampling instant, and n is data set size.
Step 6, as shown in fig. 6, building deep learning network, by training dataset be D training deep learning network, mesh
Be the mapping ruler learnt from CSI sequence to human joint points, depth network includes the AlphaPose as teacher's network
It (referred to as T ()) and is formed as student network WiSPPN (referred to as S ()), student network WiSPPN includes coding module, spy
Levy extraction module and decoder module.
Coding module is CSI-Net network, for by CSI value CtSize be upsampled to RGB standard size；
Characteristic extracting module completes feature extraction using ResNets network, by 4 basic ResNet block (16 volumes
Lamination) it is used as feature extractor, for extracting personage's posture feature, the CSI value after feature is extracted in output.
Decoder module learns personage's posture feature using FCN network method, exports the corresponding prediction square of posture adjacency matrix
Battle array pPAMt。
Teacher's network, for according to video frame ItExport posture adjacency matrix PAMt, to prediction matrix pPAMtIt optimizes.
Step 7, teacher's network export posture adjacency matrix PAMt, for each (It,Ct), video frame ItAs T ()
Input, T () export corresponding human joint points coordinate and confidence level (xt,yt；ct), posture adjacency matrix is converted by output
PAMt, t expression sampling instant；So the operation form of teacher's network is T (It)→PAMt, wherein PAMtIt is for cross-module formula prison
S () is superintended and directed, the specific method is as follows.
Step 7.1, the personage arrived using cameras capture in AlphaPose method processing step 3 are executing each human body
The picture captured is handled by person detector, obtains personage in video frame images by the picture of each video frame when posture
Bounding box, by bounding box again by posture return device handle, obtain the posture of the human body in personage's bounding box, and obtain
It is able to (xi,yi；ci) format generate n three element predictions；
Wherein, n is the quantity of the artis to be estimated, and the coordinate of artis is (x, y, c), as shown in Figure 5；In this work
N=18, xiAnd yiIt is the cross, ordinate of i-th of artis, c respectivelyiIt is the confidence level of body joint point coordinate；
Step 7.2, the posture for the human body that step 7.1 is obtained and three element predictive conversions are posture adjacency matrix PAMt,
Posture adjacency matrix PAMtIt is made of three submatrixs (x ', y ', c '), matrix x ', matrix y ' and matrix c ' pass through following formula
From 18 three element entry (xi,yi；ci), i ∈ [1,2 ... 18] is middle to be generated, and one 3 × 18 × 18 PAM has been obtainedtMatrix, i.e.,
PAM∈R3×18×18, the displacement between character joint point coordinate and artis is successfully embedded into PAMtIn.
Wherein, i is the row of matrix, and j is matrix column.
Step 8, student network WiSPPN (referred to as S ()), in the training stage, by CSI value CtAs the input of S (),
By coding module, characteristic extracting module and decoder module, the corresponding prediction matrix pPAM of posture adjacency matrix is exportedt, specifically
Method is as follows.
Step 8.1, by coding module by CSI value CtSize up-sample to the size of RGB image；
Since the sample rate of WiFi equipment and video camera is respectively set to 100Hz and 20Hz.Therefore, a paired data collection
(It,Ct) in every 5 continuous CSI values and a picture frame it is synchronous by their timestamp.According to step 4 it is found that CSI size
For m × 30 × 3 × 3, so Ct∈R5×30×3×3, C is regenerated along the time axist∈R150×3×3；But general RGB image is big
Small is 3 × 224 × 224, so needing to expand CSI value CtWidth and height with reach matching；
In order to expand CSI value CtWidth and height, CtAs input, by coding module by CSI value CtIt is upsampled to suitable
When width and height to be suitable for after characteristic extracting module, specific coding process is as follows:
CSI-Net Web vector graphic bilinear interpolation operation is directly inputted Ct∈R150×3×3It carries out being upsampled to Ct∈
R150×144×144, it is ready for next step feature extraction；
Step 8.2 will pass through the C after up-samplingtAs input, personage's Attitude estimation is learnt by characteristic extracting module
Validity feature, specific features extraction process is as follows:
As shown in fig. 7, feature extraction is completed using ResNets network, by 4 basic ResNet block (16 convolution
Layer) it is used as feature extractor, by Ct∈R150×144×144By the study of ResNets network, F is exportedt∈R300×18×18, FtBe through
CSI value after crossing feature extraction, Ct→FtSpecific conversion process is as shown in Figure 8.
Step 8.3, by the F after feature extractiontAs input, personage's posture feature is learnt by decoder module, it is defeated
The corresponding prediction pPAM of posture adjacency matrix outt；
As shown in figure 9, specific decoding process is as follows: by Ft∈R300×18×18As input, passed through using FCN network method
Two convolutional layers learn personage's posture feature, and export the corresponding prediction matrix pPAM of posture adjacency matrixt, to be reconstructed one
The spatial information of dimension；Since human joint points can be positioned as x, two coordinates of y, so output is pPAMt∈R2×18×18；
Step 10, using posture adjacency matrix PAM to pPAMtSupervision optimization, the pPAM after defeated place's optimizationt, optimization method is such as
Under；
In the training stage, by the accordingly result PAM ∈ R of teacher's network3×18×18It supervises, the result pPAM of student networkt
∈R2×18×18As prediction, by loss function L to pPAMtIt constantly optimizes, once student network learns very well, it is just
It can obtain and only use CtInput carries out single pose estimation pPAMtAbility.
Wherein, pPAMxAnd PAMxIt is the prediction and supervision of the posture adjacency matrix of human joint points coordinate in x-axis respectively,
pPAMyAnd PAMyIt is the prediction and supervision of the posture adjacency matrix of human joint points coordinate in y-axis respectively；
Step 11, student network WiSPPN input C for eacht, student network WiSPPN is made that posture is adjacent
The corresponding prediction matrix pPAM of matrixt, i.e. operation form is S (Ct)→pPAMt, so as to complete human body attitude estimation.
Fine granularity estimation method of human posture provided by the invention based on radio frequency signal, using with three antennas
Wireless transmitting terminals establish wifi, can be to wifi channel when user walks or make certain movement in the wifi area of coverage
Specific influence is generated, receive wifi signal using the wireless receiving end with three antennas and calculates the CSI value of human body attitude,
CSI is encoded, feature is extracted, decodes this sequence of operations, uses camera synchronous with the timestamp of WiFi antenna alignment
Personage's video is captured, the picture captured when moving or execute certain posture in wifi using people passes through person detecting
Device and posture return device output personage's posture coordinate and confidence level, output result are converted to posture adjacency matrix (PAM), to logical
The corresponding prediction (pPAM) for crossing personage's posture adjacency matrix that wifi signal processing obtains exercises supervision and optimizes, thus real
Single Attitude estimation is now carried out using wifi equipment, the present invention can meet convenient while realizing higher recognition accuracy
Property, ease for use and safety, and do not need user and carry any special installation, not will record the privacy life of user, there is side
Easy deployment, highly-safe feature.
As described in Figure 10, it needs to emit stable wifi using the wireless transmitting terminals with three antennas in system deployment
, and making the wireless receiving end with three antennas, the quantity for increasing antenna can capture the signal from different paths,
It can produce a variety of different Signal averaging modes at receiving antenna, be conducive to improve the identification precision to human body attitude, make
Result compared to using the equipment such as radar, camera to carry out human perception more fine granularity.
The present invention has the following advantages: compared with prior art using the human body attitude estimation technique in the present invention, this is the
One carries out the work that human perception estimates human body attitude using general commercial WiFi equipment, and work before is to utilize phase
What machine, radar or laser radar estimated human body attitude；All than radar and laser radar using general commercial WiFi equipment
Cheaper and more power saving, it is constant to illuminating, and almost without privacy concern compared with using camera to carry out human body attitude estimation.
Moreover, fine-grained human perception, energy may be implemented relative to using camera, radar or laser radar apparatus to do human perception
Enough while realizing higher recognition accuracy, meet a kind of fine granularity based on radio frequency signal of convenience and safety
Estimation method of human posture.
The above content is merely illustrative of the invention's technical idea, and this does not limit the scope of protection of the present invention, all to press
According to technical idea proposed by the present invention, any changes made on the basis of the technical scheme each falls within claims of the present invention
Protection scope within.
Claims (6)
1. a kind of fine granularity estimation method of human posture based on radio frequency signal, which comprises the following steps:
S1, using with more antennas wireless transmitting terminals and wireless receiving end establish the field wifi of waveform stabilization, while in nothing
A synchronization camera being aligned with wireless transmitting terminals timestamp is arranged in line transmitting terminal；
S2, performance objective execute human body attitude in wifi, and wireless receiving end is received across performance objective, and are executing mesh
The wifi signal for reflecting and reflecting on mark or surrounding objects, is calculated human body appearance according to the received wifi signal in wireless receiving end
The channel state information CSI value of state, meanwhile, the synchronous camera of timestamp obtains the video pictures of human body attitude, completes the receipts of data
Collection；
S3, training dataset D is established according to the picture of channel state information CSI value and all video frames；
D={ (It,Ct),t∈[1,n]}；
Wherein, ItAnd CtThe respectively video frame of synchronization and CSI value, t indicate sampling instant, and n is data set size；
S4, building deep learning network；
S5, deep learning network is trained according to training dataset D, trained deep learning network is obtained, according to instruction
The depth network perfected carries out human body attitude estimation.
2. the fine granularity estimation method of human posture based on radio frequency signal according to claim 1, which is characterized in that step
The method that channel state information CSI value is obtained in rapid S2 is specific as follows；
CSI=m × a × b × c
Wherein, m is the quantity of the WiFi packet received, and a is OFDM carrier number, and b and c are wireless transmitting terminals and wireless receiving respectively
The antenna amount at end.
3. the fine granularity estimation method of human posture based on radio frequency signal according to claim 1, which is characterized in that step
Deep learning network described in rapid S3 includes teacher network and student network, and student network includes coding module, characteristic extracting module
And decoder module；
Coding module, for by CSI value CtSize up-sample to RGB standard size；
Characteristic extracting module, for the C after up-samplingtFeature is extracted, and exports the CSI value F after extracting featuret；
Decoder module, for according to the CSI value F after extraction featuret, export the corresponding prediction matrix pPAM of posture adjacency matrixt；
Teacher's network, for according to video frame ItExport posture adjacency matrix PAMt, to prediction matrix pPAMtExercise supervision optimization.
4. the fine granularity estimation method of human posture based on radio frequency signal according to claim 3, which is characterized in that institute
State teacher's network output posture adjacency matrix PAMtMethod it is as follows:
Firstly, obtaining the boundary of performance objective in video frame using each frame picture in AlphaPose method processing video
Frame, then bounding box is returned into device by posture and is handled, the posture of performance objective in bounding box is obtained, and generate n three
Element prediction, n are the quantity of the artis to be estimated；
Then, by the posture of human body and three element predictive conversion posture adjacency matrix PAMt。
5. the fine granularity estimation method of human posture based on radio frequency signal according to claim 4, which is characterized in that institute
State student network output posture adjacency matrix pPAMtMethod it is as follows:
The coding module is CSI-Net network, using bilinear interpolation operation by Ct∈R150×3×3It carries out being upsampled to Ct∈
R150×144×144；
Characteristic extracting module is 4 basic ResNet blocks in ResNets network, Ct∈R150×144×144By ResNets network
Study, export Ft∈R300×18×18；
Decoder module is FCN network, inputs the CSI value F after feature extractiont, special by two convolutional layer study personage's postures
Sign exports the corresponding prediction matrix pPAM of posture adjacency matrixt。
6. the fine granularity estimation method of human posture based on radio frequency signal according to claim 5, which is characterized in that adopt
With loss function L to prediction matrix pPAMtIt optimizes, specific as follows:
Wherein, pPAMxAnd PAMxIt is the prediction matrix of human joint points coordinate and posture adjacency matrix, pPAM in x-axis respectivelyyWith
PAMyIt is the prediction matrix of human joint points coordinate and posture adjacency matrix in y-axis respectively.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910457667.7A CN110245588A (en) | 2019-05-29 | 2019-05-29 | A kind of fine granularity estimation method of human posture based on radio frequency signal |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910457667.7A CN110245588A (en) | 2019-05-29 | 2019-05-29 | A kind of fine granularity estimation method of human posture based on radio frequency signal |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110245588A true CN110245588A (en) | 2019-09-17 |
Family
ID=67885390
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910457667.7A Pending CN110245588A (en) | 2019-05-29 | 2019-05-29 | A kind of fine granularity estimation method of human posture based on radio frequency signal |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110245588A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110728213A (en) * | 2019-09-26 | 2020-01-24 | 浙江大学 | Fine-grained human body posture estimation method based on wireless radio frequency signals |
| CN110737201A (en) * | 2019-10-11 | 2020-01-31 | 珠海格力电器股份有限公司 | monitoring method, device, storage medium and air conditioner |
| CN111369618A (en) * | 2020-02-20 | 2020-07-03 | 清华大学 | Human body posture estimation method and device based on compressed sampling RF signals |
| CN112446923A (en) * | 2020-11-23 | 2021-03-05 | 中国科学技术大学 | Human body three-dimensional posture estimation method and device, electronic equipment and storage medium |
| CN112668439A (en) * | 2020-12-23 | 2021-04-16 | 中国科学技术大学 | Human body posture estimation method and device based on wireless signals |
| CN112668413A (en) * | 2020-12-16 | 2021-04-16 | 北京邮电大学 | Human body posture estimation method and device, electronic equipment and readable storage medium |
| CN112911267A (en) * | 2021-01-29 | 2021-06-04 | 北京邮电大学 | Indoor imaging method and device based on WiFi channel state information |
| CN114246582A (en) * | 2021-12-20 | 2022-03-29 | 杭州慧光健康科技有限公司 | System and method for detecting bedridden people based on long-term and short-term memory neural network |
| CN115034259A (en) * | 2022-05-17 | 2022-09-09 | 合肥工业大学 | Human body action identification method based on position independence of CSI (channel state information) and double-thread convolution network |
| CN115412188A (en) * | 2022-08-26 | 2022-11-29 | 福州大学 | Power distribution station room operation behavior identification method based on wireless sensing |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140031980A1 (en) * | 2011-11-11 | 2014-01-30 | Jens-Steffen Gutmann | Systems and methods for extending slam to multiple regions |
| CN106066995A (en) * | 2016-05-25 | 2016-11-02 | 西安交通大学 | A kind of wireless unbundling human body behavioral value algorithm |
| CN106780569A (en) * | 2016-11-18 | 2017-05-31 | 深圳市唯特视科技有限公司 | A kind of human body attitude estimates behavior analysis method |
| CN107180223A (en) * | 2017-04-10 | 2017-09-19 | 南京苗米科技有限公司 | Action identification method and system based on WIFI wireless signals |
-
2019
- 2019-05-29 CN CN201910457667.7A patent/CN110245588A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20140031980A1 (en) * | 2011-11-11 | 2014-01-30 | Jens-Steffen Gutmann | Systems and methods for extending slam to multiple regions |
| CN106066995A (en) * | 2016-05-25 | 2016-11-02 | 西安交通大学 | A kind of wireless unbundling human body behavioral value algorithm |
| CN106780569A (en) * | 2016-11-18 | 2017-05-31 | 深圳市唯特视科技有限公司 | A kind of human body attitude estimates behavior analysis method |
| CN107180223A (en) * | 2017-04-10 | 2017-09-19 | 南京苗米科技有限公司 | Action identification method and system based on WIFI wireless signals |
Non-Patent Citations (1)
| Title |
|---|
| FEI WANG 等: ""CanWiFi Estimate Person Pose？"", 《ARXIV:1904.00277V2 [CS.CV]》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110728213A (en) * | 2019-09-26 | 2020-01-24 | 浙江大学 | Fine-grained human body posture estimation method based on wireless radio frequency signals |
| CN110737201A (en) * | 2019-10-11 | 2020-01-31 | 珠海格力电器股份有限公司 | monitoring method, device, storage medium and air conditioner |
| CN111369618A (en) * | 2020-02-20 | 2020-07-03 | 清华大学 | Human body posture estimation method and device based on compressed sampling RF signals |
| CN112446923A (en) * | 2020-11-23 | 2021-03-05 | 中国科学技术大学 | Human body three-dimensional posture estimation method and device, electronic equipment and storage medium |
| CN112668413A (en) * | 2020-12-16 | 2021-04-16 | 北京邮电大学 | Human body posture estimation method and device, electronic equipment and readable storage medium |
| CN112668413B (en) * | 2020-12-16 | 2022-11-08 | 北京邮电大学 | Human body posture estimation method and device, electronic equipment and readable storage medium |
| CN112668439A (en) * | 2020-12-23 | 2021-04-16 | 中国科学技术大学 | Human body posture estimation method and device based on wireless signals |
| CN112911267A (en) * | 2021-01-29 | 2021-06-04 | 北京邮电大学 | Indoor imaging method and device based on WiFi channel state information |
| CN114246582A (en) * | 2021-12-20 | 2022-03-29 | 杭州慧光健康科技有限公司 | System and method for detecting bedridden people based on long-term and short-term memory neural network |
| CN115034259A (en) * | 2022-05-17 | 2022-09-09 | 合肥工业大学 | Human body action identification method based on position independence of CSI (channel state information) and double-thread convolution network |
| CN115034259B (en) * | 2022-05-17 | 2024-02-20 | 合肥工业大学 | Human body action recognition method based on CSI and irrelevant to position of double-wire Cheng Juanji network |
| CN115412188A (en) * | 2022-08-26 | 2022-11-29 | 福州大学 | Power distribution station room operation behavior identification method based on wireless sensing |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110245588A (en) | A kind of fine granularity estimation method of human posture based on radio frequency signal | |
| CN106598226B (en) | A kind of unmanned plane man-machine interaction method based on binocular vision and deep learning | |
| CN110728213A (en) | Fine-grained human body posture estimation method based on wireless radio frequency signals | |
| CN103136986B (en) | Sign Language Recognition Method and system | |
| Li et al. | Unsupervised learning for human sensing using radio signals | |
| CN108898063A (en) | A kind of human body attitude identification device and method based on full convolutional neural networks | |
| WO2021190321A1 (en) | Image processing method and device | |
| CN111582200B (en) | Human body posture estimation method, device, electronic equipment and medium | |
| CN111914615A (en) | Fire-fighting area passability analysis system based on stereoscopic vision | |
| CN114241379A (en) | Passenger abnormal behavior identification method, device and equipment and passenger monitoring system | |
| CN112950673A (en) | Target object detection and tracking system and method | |
| CN108460794A (en) | A kind of infrared well-marked target detection method of binocular solid and system | |
| CN116194951A (en) | Method and apparatus for stereoscopic based 3D object detection and segmentation | |
| CN113935379B (en) | Human body activity segmentation method and system based on millimeter wave radar signals | |
| Dai et al. | Indoor 3D human trajectory reconstruction using surveillance camera videos and point clouds | |
| CN113050083A (en) | Ultra-wideband radar human body posture reconstruction method based on point cloud | |
| Wang et al. | Real-time Through-Wall Multi-Human Localization and Behavior Recognition Based on MIMO Radar | |
| CN111656356A (en) | Object recognition system using distributed neural network | |
| Xie et al. | RPM: RF-based pose machines | |
| Xie et al. | Accurate human pose estimation using rf signals | |
| CN203397395U (en) | Moving object detection device in platform based on DSP + FPGA | |
| CN111542006A (en) | Object identification method based on wireless radio frequency signal | |
| Zhang et al. | Cross-domain gesture recognition via learning spatiotemporal features in Wi-Fi sensing | |
| CN116964599A (en) | Three-dimensional human body motion capturing method, training method and system based on laser radar | |
| CN109215122A (en) | A kind of streetscape three-dimensional reconstruction system and method, intelligent carriage |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190917 |