CN109902558A - A kind of human health deep learning prediction technique based on CNN-LSTM - Google Patents
A kind of human health deep learning prediction technique based on CNN-LSTM Download PDFInfo
- Publication number
- CN109902558A CN109902558A CN201910035483.1A CN201910035483A CN109902558A CN 109902558 A CN109902558 A CN 109902558A CN 201910035483 A CN201910035483 A CN 201910035483A CN 109902558 A CN109902558 A CN 109902558A
- Authority
- CN
- China
- Prior art keywords
- gait
- cnn
- pedestrian
- feature
- lstm
- 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.)
- Granted
Links
Landscapes
- Image Analysis (AREA)
Abstract
The human health deep learning prediction technique based on CNN-LSTM that the present invention relates to a kind of by the pathology gait of advanced deep learning models coupling camera pedestrian for identification, and then predicts its health status.Dynamical fashion is used to the identification of pedestrian's gait, the gait feature of pedestrian in video image is extracted with CNN, is then based on chronicle recognition gait feature using LSTM, and then differentiate its gait types.To realize more people's on-line real-time measuremens, video pedestrian's detection module is added in systems, and constructs a light weight and the strong composite CNN model of Feature Mapping ability, both the identification for pedestrian in video detection, was also used for the feature extraction of pedestrian's gait.The present invention is based on images to identify gait using dynamical fashion, and hardware realization is simple, but technologically advanced, high to the recognition accuracy of gait, has a extensive future, can be used to region measuring of human health and analysis, it can also be used to which customizing service is accomplished by family.
Description
Technical field
The present invention relates to deep learning and human health field, specifically a kind of human health based on CNN-LSTM is deep
Degree study prediction technique.
Background technique
With the development of economy, people's lives level is greatly improved, while people also get over the concern of health
Come higher；Human health status is monitored, and the disease that human body is likely to occur is made and is forecast in advance, is had in control and prevention of disease field
Significance.Studies have shown that gait and its health status on foot of people are closely bound up, pathology gait common at present has scissors
Step, mop step, drunk step, hemiplegic gait, paraparetic gait, gluteus maximums gait, glutaeus medius gait, quadriceps femoris gait, cross-domain step
State, arthrocleisis gait, reeling gait, festinating gait, short-leg gait, painful gait etc. correspond to it not without synchronous state again
Same pathology, such as: scissors step, both feet are separated, and the rear foot is toward former riding, when front foot keeps separating just now toward former riding again and the rear foot
Distance mostly occurs in the patient of cerebrovascular disease, cerebral infarction, cerebral hemorrhage or spinal cord injury；Mop step, this kind of patient walk
When one leg be normal, but due to muscle or the obstacle of nervous function, another one leg can be pulled through at leisure to be come；It is drunk
Step, which usually walks less than on straight line, vacillating and staggering, have the people of this gait answer strong suspicion brain tumor, cerebral hemorrhage,
Lesions on cerebellum etc..
Human body is walked the variation of gait slowly all normal degenerative changes at last, and normal scope is also calculated.If
It is to occur the exception of some gaits in the short time rapidly, and development speed is very fast, then with other some nervous systems
Symptom, this when will strong suspicion be that some system there is a problem.Based on the above, the present invention proposes that one kind is based on
The human health deep learning prediction technique of CNN-LSTM.Convolutional neural networks (Convolutional Neural
Networks, CNN) it is deep learning field important algorithm, it is mainly used for the feature extraction or classification of image, a Zhang Erwei or more
The digital picture of higher-dimension can be directly as its input.CNN algorithm had excellent ability in feature extraction, in the world in 2015
In ImageNet contest, the image recognition error rate of the deep learning model based on CNN algorithm is lower than the mankind down to 4.94%
5.1% error rate.Shot and long term memory network (Long Short-Term Memory, LSTM) is mainly for the treatment of time sequence
Column data is widely applied to the fields such as speech recognition, text identification.The present invention proposes a kind of human body based on CNN-LSTM
Healthy deep learning prediction technique constructs a light weight quickly and the composite CNN model with stronger Feature Mapping ability, uses
The identification and Method of Gait Feature Extraction of pedestrian in image, and the gait feature of extraction is inputted into LSTM network, then LSTM is based on
Timing identifies gait feature using dynamic mode, differentiates pedestrian's gait types, and then predict pedestrian's health status.
Summary of the invention
The human health deep learning prediction technique based on CNN-LSTM that the object of the present invention is to provide a kind of；The invention is logical
Cross camera and obtain human body and walk the image sequence of gait, then extract the gait feature in every image using CNN, and by its
Human health status is prejudged, and make by the variation of the gait feature in LSTM algorithm foundation image sequence as the input of LSTM
It forecasts in advance out.Its major advantage has:
1, the deep learning model in forward position is used for human health prediction, not only can recognize that the pedestrian in video, but also
More people's gaits can also be detected simultaneously；Its hardware device is simple, and application prospect is wide, can be used for the strong of crowd under specific environment
Health state analysis, it can also be used to which customizing service is accomplished by family；
2, the deep learning model that designs of the present invention, light weight and has stronger Feature Mapping ability, extracts characteristics of image
Speed is fast, is able to achieve more people in video and walks the real time discriminating of gait；
The present invention realizes that goal of the invention adopts the following technical scheme that
A kind of human health deep learning prediction technique based on CNN-LSTM mainly includes hardware and software two parts,
Hardware is camera and computing module, and software mainly includes three modules, is pedestrian detection module in image respectively, is based on
The characteristic extracting module of CNN and characteristic sequence identification module based on LSTM.User is obtained first with camera to walk video,
Secondly every image in preprocessed video (be sized, choose clear image, etc.), then each row in detection image
People, and each pedestrian is extracted, then using the pedestrian image of extraction as the input of CNN model, the gait feature of pedestrian is obtained,
Finally using gait feature sequence as the input of LSTM, human health status is differentiated.For customizing user, obtains and walk in CNN
State feature stage can also identify user, combine its gait feature to establish state of health data library user information, realize long-term inspection
It surveys, by analyzing the situation of change of its gait, provides more accurate health status diagnosis, while it is pre- to be also able to achieve user health
It surveys.
Preferably, a kind of human health deep learning prediction technique based on CNN-LSTM provided by the invention, it will be first
Into deep learning models coupling camera, by analyze user walk when gait, differentiate the health status of user.
Preferably, a kind of human health deep learning prediction technique based on CNN-LSTM provided by the invention, building
One light weight and the stronger CNN model of Feature Mapping ability are used to extract pedestrian's gait feature in image, by the feature of extraction
LSTM is inputted, pedestrian's gait is identified using dynamic mode.
Preferably, a kind of human health deep learning prediction technique based on CNN-LSTM provided by the invention, passes through
Pedestrian detection technology obtains each pedestrian image in video, CNN Feature Selection Model is sequentially input, to realize in video
More people diagnose simultaneously.
Preferably, a kind of human health deep learning prediction technique based on CNN-LSTM provided by the invention, for
Domestic consumer identifies user, it can be achieved that customizing service while feature extraction, then combines its gait to build user information
Vertical database, analyzes user's gait feature and gait situation of change, provides more accurate Gernral Check-up and prediction.
Preferably, a kind of human health deep learning prediction technique based on CNN-LSTM provided by the invention, video
Middle pedestrian detection uses Faster-RCNN technology, realizes the quick detection of pedestrian；Before pedestrian detection, in preprocessed video
Image abandons fuzzy image, to improve discrimination.
Preferably, a kind of human health deep learning prediction technique based on CNN-LSTM provided by the invention, is real
Now to pedestrian's gait feature extract real-time in video and with stronger Feature Mapping ability, building one contains 11 convolutional layers
Light weight CNN model, the pixel placement of input picture is 120*120, is maximum quickenings feature extraction speed, selection multistep
Long convolutional layer replaces pond layer.
The utility model has the advantages that
Compared with prior art, the present invention its advantages are embodied in: by advanced deep learning models coupling camera
For predicting human health status, hardware realization is simple, but technologically advanced, and application prospect is wider.By CNN combination LSTM in system
The Dynamic Recognition for realizing pedestrian's gait is identified, the Dynamic Recognition mode based on sequence compared to based on single image static state gait feature
With higher discrimination；By video pedestrian detection technology, it is able to achieve multirow people real-time detection in video, in conjunction with big data skill
Art is able to achieve the assessment of certain region population health and prediction；It for domestic consumer, can accomplish customizing service, extract and walk in CNN
User is identified while state feature, then combines its gait feature to establish state of health data library user information, by long-term
Observation, analyzes the situation of change of its gait, provides more accurate health status diagnosis, while making to its health status pre-
It surveys；The present invention designs a kind of lightweight composite CNN model, has stronger Feature Mapping ability, and speed is fast, is able to achieve reality
When feature extraction.
Detailed description of the invention
Fig. 1 system construction drawing
Fig. 2 pedestrian detection process
The light-type CNN model structure schematic diagram that Fig. 3 present invention constructs
Gait Recognition schematic diagram of the Fig. 4 based on LSTM
The convergence speed comparison diagram of the light-type CNN of Fig. 5 building
Specific embodiment:
A kind of human health deep learning prediction technique based on CNN-LSTM, realization process such as Fig. 1 show, first with
Camera obtains user and walks video, secondly every image in preprocessed video (be sized, choose clear image, etc.),
Then each pedestrian in detection image, and each pedestrian is extracted, then using the pedestrian image of extraction as the defeated of CNN model
Enter, and then obtain the gait feature of pedestrian, the gait feature sequence being made of plurality of pictures that finally will acquire is as LSTM's
Input, differentiates its gait types of walking, to predict human health status；For customizing user, gait feature is obtained in CNN
Stage can also identify user, combine its gait feature to establish state of health data library user information, realize long-term detection, mentioned
It is predicted for more effective health status.
The present invention uses dynamical fashion to the identification of pedestrian's gait, i.e., step when detected person walks is obtained using camera
State video differentiates its gait of walking, and then analyze its health status by a series of images of walking of analysis detected person.To protect
The discrimination for hindering gait first has to screen the image that camera obtains, and except the picture of deblurring, it is big secondly to adjust image
It is small, image pedestrian's detection is then carried out, finds out each pedestrian in image, and then differentiate the gait of each pedestrian.Due to this hair
Bright is its health status to be differentiated based on gait, and the speed of walking of people slows down relatively, and for the calculation amount for reducing hardware, camera shooting is obtained
Video image, therefrom choose 5 each second.Explanation is further explained to the present invention below by way of specific embodiment.
1, image procossing and screening
Screening for image, generally need to be to image preprocessing, and then clarity differentiates, finally selects clear picture.?
Under the premise of the image of high quality is selected in guarantee, to reduce calculation amount as far as possible, image pretreatment operation of the invention includes figure
As size adjusting, Gauss denoising and gray processing.
The clarity of image generally indicates that gradient is bigger with gradient, illustrates that the edge of image is more clear, on the contrary then mould
Paste.The present invention selects Tenengrad function to judge the clarity of image.Tenengrad function uses Sobel operator extraction
Gradient value both horizontally and vertically is worth bigger, representative image by the average gray value of Sobel operator treated image
It is more clear.If GxAnd GyFor the convolution kernel of Sobel, I (x, y) is pixel, the calculating of Tenengrad gradient function as shown in formula:
2, pedestrian detection
Along with the fast development of depth, image detecting technique is also more and more mature, the mould currently used for image detection
Type has RCNN, Fast-RCNN and Faster-RCNN etc., wherein and the network structure of Faster-RCNN is the most complicated, but it
Image detection speed and Detection accuracy are also highest.The realization of Faster-RCNN is broadly divided into three processes:
Firstly, input picture passes through the processing of pre-training CNN model, convolution characteristic pattern (conv feature is obtained
map)。
Then, RPN (Region Propose Network) handles the convolution characteristic pattern of extraction, and RPN is for seeking
Looking for may be comprising the region (regions, bounding box) of the predefined quantity of target.Comprising two parts, one for predicting target area
The center anchor point in domain corresponding coordinate x, y and width high w, h；Second is that for determining that the region is prospect or background.
Finally, being based on R-CNN module, classify to the content in bounding box, adjusts bounding box coordinates.
For the real-time for guaranteeing system, Faster-RCNN model is used for the pedestrian detection in image by the present invention, and is improved
The composite CNN of building is used for the feature extraction in its first step, realizes process such as by traditional Faster-RCNN model
Shown in Fig. 3.
3, CNN feature extraction
CNN has excellent performance to image characteristics extraction or identification, studies have shown that the network of CNN is deeper, characterology
Habit ability is stronger, but with the intensification of network, what network became is increasingly difficult to train, and the calculation amount of model also becomes increasingly
Greatly.In convolutional neural networks, convolutional layer is used for feature extraction, and convolution kernel is bigger, and the local correlation information of extraction is more, pond layer
It is mainly used for the dimensionality reduction of characteristic pattern, the distortion and rotation to image are insensitive, and batch normalized essence is the normalization of data.
Different layer building depth networks is exactly selected in the building of CNN model according to actual demand.CNN is used for picture in the present invention
The extraction of middle gait feature and pedestrian's identification, construct a kind of light weight and the stronger compound CNN model of Feature Mapping ability, can
Gait feature identification also may be implemented in the identification for realizing pedestrian, specific as shown in Figure 3.
In Fig. 3, ' ker ' represents convolution kernel size, and ' str ' is convolution step-length, and ' B_Norm ' is batch normalization layer.To realize
The lightweight of model, 11 convolutional layers of CNN model selection of building, 6 batches of normalization layers.The present invention has inputted the picture of CNN
By pre-processing and screening, there is no rotation and distortion problems, so pond layer is replaced with the long convolutional layer of multistep, to reduce model
Calculation amount.For the convergence rate for improving model, the composite CNN model that the present invention constructs is connected using local residual error, and
It is all connected with one batch of normalization layer after each residual error layer, is specifically shown in Fig. 3.For the calculation amount for reducing LSTM, while obtaining more effective
Gait feature, second of CNN model is complete, and articulamentum is set as 32.
4, based on the Gait Recognition of LSTM
LSTM is to solve the long more effective technology of sequence Dependence Problem, and be widely used in natural language processing and voice
Identification.The improved core of LSTM is the cell that a discriminant information is added and whether is remembered, in LSTM algorithm, each cell
Door there are three all connecting, respectively forgetting door, input gate and out gate, as shown in Figure 4.F, g and h is activation primitive.Information exists
In LSTM communication process, the output of Input Gate such as formula 2:
In formula 2, xtFor the input of current time Input Gate, st-1For all cell in the last moment block
Output, bt-1For the output of other block of last moment.The output of Forget Gate such as formula 3:
The output of Output Gate such as formula 4:
The output of Cell such as formula 5:
In the present invention, the output for the gait feature that CNN Feature Selection Model obtains is set as 32, so the input of LSTM
Width is set as 32.
Experimental verification part
For the convergence rate for the composite CNN model that the verifying present invention constructs, it is used for cifar-10 common data sets
The step-length that convolution kernel is 5 is set as 1, other are constant since the resolution ratio of the picture of cifar-10 is 32*32 by classification；In addition,
The famous deep learning model Vgg for there are 11 convolutional layers is also used for the classification of cifar-10 data, result is right therewith
Than.It include 5 pond layers since the input of Vgg model is 224*224；It, will for the size for adapting to cifar-10 image data
Vgg-11 the first two pond layer, and the convolution kernel of first convolutional layer is set as 3*3.Convergence rate when two model trainings is as schemed
Shown in 5.
In Fig. 5, ' Vgg-11 ' is training error curve of the classic network Vgg-11 on cifar-10, and ' My-CNN ' is this
Error curve of the light weight CNN model on cifar-10 for inventing building, from the comparison of the error curve of two models it is found that
The error convergence speed for the light-type CNN that the present invention constructs is faster.In summary, the light-type CNN that the present invention constructs not only has
There is rapidity, and there is stronger Feature Mapping ability.
It is obvious to a person skilled in the art that invention is not limited to the details of the above exemplary embodiments, Er Qie
In the case where without departing substantially from spirit or essential attributes of the invention, the present invention can be realized in other specific forms.Therefore, no matter
From the point of view of which point, the present embodiments are to be considered as illustrative and not restrictive, and the scope of the present invention is by appended power
Benefit requires rather than above description limits, it is intended that all by what is fallen within the meaning and scope of the equivalent elements of the claims
Variation is included within the present invention.Any reference signs in the claims should not be construed as limiting the involved claims.
In addition, it should be understood that although this specification is described in terms of embodiments, but not each embodiment is only wrapped
Containing an independent technical solution, this description of the specification is merely for the sake of clarity, and those skilled in the art should
It considers the specification as a whole, the technical solutions in the various embodiments may also be suitably combined, forms those skilled in the art
The other embodiments being understood that.
Claims (7)
1. a kind of human health deep learning prediction technique based on CNN-LSTM, it is characterised in that: main including hardware and soft
Part two parts, hardware are camera and computing module, and it is pedestrian detection mould in image respectively that software, which mainly includes three modules,
Block, the characteristic extracting module based on CNN and the characteristic sequence identification module based on LSTM.User is obtained first with camera to walk
Road video, secondly every image in preprocessed video (be sized, choose clear image, etc.), then in detection image
Each pedestrian, and each pedestrian is extracted, then using the pedestrian image of extraction as the input of CNN model, obtain the step of pedestrian
State feature differentiates human health status finally using gait feature sequence as the input of LSTM.For customizing user, in CNN
User can also be identified by obtaining the gait feature stage, combined its gait feature to establish state of health data library user information, realized
Long-term detection, by analyzing the situation of change of its gait, provides more accurate health status diagnosis, while being also able to achieve user
Health forecast.
2. a kind of human health deep learning prediction technique based on CNN-LSTM according to claim 1, feature exist
In: by advanced deep learning models coupling camera, gait when being walked by analyzing user differentiates the healthy shape of user
Condition.
3. a kind of human health deep learning prediction technique based on CNN-LSTM according to claim 1, feature exist
In: it constructs a light weight and the stronger CNN model of Feature Mapping ability is used to extract pedestrian's gait feature in image, will extract
Feature input LSTM, the health status of user is differentiated using dynamic mode.
4. a kind of human health deep learning prediction technique based on CNN-LSTM according to claim 1, feature exist
In: by pedestrian detection technology, each pedestrian image in video is obtained, CNN Feature Selection Model is sequentially input, to realize
More people diagnose simultaneously in video.
5. a kind of human health deep learning prediction technique based on CNN-LSTM according to claim 1, feature exist
In: for domestic consumer, it can be achieved that customizing service, user is identified while feature extraction, then combines user information
Its gait establishes database, analyzes user's gait feature and gait situation of change, provides more accurate Gernral Check-up and prediction.
6. a kind of human health deep learning prediction technique based on CNN-LSTM according to claim 1, feature exist
In: pedestrian detection uses Faster-RCNN technology in video, realizes the quick detection of pedestrian；Before pedestrian detection, pretreatment view
Image in frequency abandons fuzzy image, to improve discrimination.
7. a kind of human health deep learning prediction technique based on CNN-LSTM according to claim 1, feature exist
In: to realize to pedestrian's gait feature extract real-time in video and with stronger Feature Mapping ability, constructs one and contain 11
The light weight CNN model of a convolutional layer, the pixel placement of input picture are 120*120, are maximum quickening feature extraction speed,
The long convolutional layer of multistep is selected to replace pond layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910035483.1A CN109902558B (en) | 2019-01-15 | 2019-01-15 | CNN-LSTM-based human health deep learning prediction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910035483.1A CN109902558B (en) | 2019-01-15 | 2019-01-15 | CNN-LSTM-based human health deep learning prediction method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109902558A true CN109902558A (en) | 2019-06-18 |
| CN109902558B CN109902558B (en) | 2022-12-20 |
Family
ID=66943643
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910035483.1A Active CN109902558B (en) | 2019-01-15 | 2019-01-15 | CNN-LSTM-based human health deep learning prediction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109902558B (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110390298A (en) * | 2019-07-23 | 2019-10-29 | 曲彦隆 | A kind of gait simulation and forecast system and simulating and predicting method |
| CN110956111A (en) * | 2019-11-22 | 2020-04-03 | 苏州闪驰数控系统集成有限公司 | Artificial intelligence CNN, LSTM neural network gait recognition system |
| CN111027417A (en) * | 2019-11-21 | 2020-04-17 | 复旦大学 | Gait recognition method and gait evaluation system based on human body key point detection algorithm |
| CN111310596A (en) * | 2020-01-20 | 2020-06-19 | 北京海益同展信息科技有限公司 | Animal diseased state monitoring system and method |
| CN111325865A (en) * | 2020-03-20 | 2020-06-23 | 广州美电恩智电子科技有限公司 | Non-inductive attendance checking method and device and equipment |
| CN111986801A (en) * | 2020-07-14 | 2020-11-24 | 珠海中科先进技术研究院有限公司 | Rehabilitation evaluation method, device and medium based on deep learning |
| CN112185558A (en) * | 2020-09-22 | 2021-01-05 | 珠海中科先进技术研究院有限公司 | Mental health and rehabilitation evaluation method, device and medium based on deep learning |
| CN112735609A (en) * | 2020-12-31 | 2021-04-30 | 浙江禾连网络科技有限公司 | LSTM model-based old people biological signal health monitoring method |
| CN112800908A (en) * | 2021-01-19 | 2021-05-14 | 中国科学院心理研究所 | Method for establishing anxiety perception model based on individual gait analysis in video |
| WO2021217906A1 (en) * | 2020-04-28 | 2021-11-04 | 深圳壹账通智能科技有限公司 | Posture detection method, apparatus and device based on gait features, and storage medium |
| CN117314926A (en) * | 2023-11-30 | 2023-12-29 | 湖南大学 | Method, apparatus and storage medium for confirming maintenance of laser modification processing apparatus |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016065534A1 (en) * | 2014-10-28 | 2016-05-06 | 中国科学院自动化研究所 | Deep learning-based gait recognition method |
| CN108090439A (en) * | 2017-12-14 | 2018-05-29 | 合肥寰景信息技术有限公司 | Pedestrian's feature extraction and processing system based on deep learning |
| CN108229390A (en) * | 2018-01-02 | 2018-06-29 | 济南中维世纪科技有限公司 | Rapid pedestrian detection method based on deep learning |
-
2019
- 2019-01-15 CN CN201910035483.1A patent/CN109902558B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016065534A1 (en) * | 2014-10-28 | 2016-05-06 | 中国科学院自动化研究所 | Deep learning-based gait recognition method |
| CN108090439A (en) * | 2017-12-14 | 2018-05-29 | 合肥寰景信息技术有限公司 | Pedestrian's feature extraction and processing system based on deep learning |
| CN108229390A (en) * | 2018-01-02 | 2018-06-29 | 济南中维世纪科技有限公司 | Rapid pedestrian detection method based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| 刘安祺等: "基于深度学习的实时人体动作识别监控方法", 《黑龙江科学》 * |
| 衣世东: "基于深度学习的图像识别算法研究", 《网络安全技术与应用》 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110390298A (en) * | 2019-07-23 | 2019-10-29 | 曲彦隆 | A kind of gait simulation and forecast system and simulating and predicting method |
| CN110390298B (en) * | 2019-07-23 | 2022-01-11 | 曲彦隆 | Gait simulation prediction system and simulation prediction method |
| CN111027417A (en) * | 2019-11-21 | 2020-04-17 | 复旦大学 | Gait recognition method and gait evaluation system based on human body key point detection algorithm |
| CN111027417B (en) * | 2019-11-21 | 2023-09-01 | 复旦大学 | Gait recognition method and gait evaluation system based on human body key point detection algorithm |
| CN110956111A (en) * | 2019-11-22 | 2020-04-03 | 苏州闪驰数控系统集成有限公司 | Artificial intelligence CNN, LSTM neural network gait recognition system |
| CN111310596A (en) * | 2020-01-20 | 2020-06-19 | 北京海益同展信息科技有限公司 | Animal diseased state monitoring system and method |
| CN111325865A (en) * | 2020-03-20 | 2020-06-23 | 广州美电恩智电子科技有限公司 | Non-inductive attendance checking method and device and equipment |
| WO2021217906A1 (en) * | 2020-04-28 | 2021-11-04 | 深圳壹账通智能科技有限公司 | Posture detection method, apparatus and device based on gait features, and storage medium |
| CN111986801A (en) * | 2020-07-14 | 2020-11-24 | 珠海中科先进技术研究院有限公司 | Rehabilitation evaluation method, device and medium based on deep learning |
| CN112185558A (en) * | 2020-09-22 | 2021-01-05 | 珠海中科先进技术研究院有限公司 | Mental health and rehabilitation evaluation method, device and medium based on deep learning |
| CN112735609A (en) * | 2020-12-31 | 2021-04-30 | 浙江禾连网络科技有限公司 | LSTM model-based old people biological signal health monitoring method |
| CN112800908A (en) * | 2021-01-19 | 2021-05-14 | 中国科学院心理研究所 | Method for establishing anxiety perception model based on individual gait analysis in video |
| CN112800908B (en) * | 2021-01-19 | 2024-03-26 | 中国科学院心理研究所 | Method for establishing anxiety perception model based on individual gait analysis in video |
| CN117314926A (en) * | 2023-11-30 | 2023-12-29 | 湖南大学 | Method, apparatus and storage medium for confirming maintenance of laser modification processing apparatus |
| CN117314926B (en) * | 2023-11-30 | 2024-01-30 | 湖南大学 | Method, apparatus and storage medium for confirming maintenance of laser modification processing apparatus |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109902558B (en) | 2022-12-20 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109902558A (en) | A kind of human health deep learning prediction technique based on CNN-LSTM | |
| Liao et al. | Deep facial spatiotemporal network for engagement prediction in online learning | |
| Bak et al. | Spatio-temporal saliency networks for dynamic saliency prediction | |
| CN106446930B (en) | Robot operative scenario recognition methods based on deep layer convolutional neural networks | |
| CN104063719B (en) | Pedestrian detection method and device based on depth convolutional network | |
| CN109815826B (en) | Method and device for generating face attribute model | |
| CN110458038B (en) | Small data cross-domain action identification method based on double-chain deep double-current network | |
| WO2020114118A1 (en) | Facial attribute identification method and device, storage medium and processor | |
| CN109902546A (en) | Face identification method, device and computer-readable medium | |
| CN109101865A (en) | A kind of recognition methods again of the pedestrian based on deep learning | |
| CN106599800A (en) | Face micro-expression recognition method based on deep learning | |
| CN110119672A (en) | A kind of embedded fatigue state detection system and method | |
| CN109543602A (en) | A kind of recognition methods again of the pedestrian based on multi-view image feature decomposition | |
| CN109543632A (en) | A kind of deep layer network pedestrian detection method based on the guidance of shallow-layer Fusion Features | |
| CN109710804B (en) | Teaching video image knowledge point dimension reduction analysis method | |
| CN109903339B (en) | Video group figure positioning detection method based on multi-dimensional fusion features | |
| CN106909938A (en) | Viewing angle independence Activity recognition method based on deep learning network | |
| CN113963032A (en) | Twin network structure target tracking method fusing target re-identification | |
| Sun et al. | Noise-tolerant RGB-D feature fusion network for outdoor fruit detection | |
| CN109165658A (en) | A kind of strong negative sample underwater target detection method based on Faster-RCNN | |
| Yang et al. | A Face Detection Method Based on Skin Color Model and Improved AdaBoost Algorithm. | |
| CN106529441A (en) | Fuzzy boundary fragmentation-based depth motion map human body action recognition method | |
| CN115346272A (en) | Real-time tumble detection method based on depth image sequence | |
| Sakthimohan et al. | Detection and Recognition of Face Using Deep Learning | |
| CN112488165A (en) | Infrared pedestrian identification method and system based on deep learning model |
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 |