CN109902558B - CNN-LSTM-based human health deep learning prediction method - Google Patents
CNN-LSTM-based human health deep learning prediction method Download PDFInfo
- Publication number
- CN109902558B CN109902558B CN201910035483.1A CN201910035483A CN109902558B CN 109902558 B CN109902558 B CN 109902558B CN 201910035483 A CN201910035483 A CN 201910035483A CN 109902558 B CN109902558 B CN 109902558B
- Authority
- CN
- China
- Prior art keywords
- gait
- cnn
- pedestrian
- lstm
- image
- 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.)
- Active
Links
Images
Abstract
The invention relates to a CNN-LSTM-based human health deep learning prediction method, which is used for predicting the health state of a pedestrian by combining an advanced deep learning model with a camera to identify the pathological gait of the pedestrian. The gait of the pedestrian is identified in a dynamic mode, the gait features of the pedestrian in the video image are extracted through CNN, and then the gait features are identified through LSTM based on time sequence, so that the gait type of the pedestrian is further judged. In order to realize online real-time detection of multiple people, a video pedestrian detection module is added into the system, and a lightweight multiplexing CNN model with strong feature mapping capability is constructed, so that the system is used for identifying pedestrians in video detection and extracting features of gait of the pedestrians. The gait recognition method based on the image adopts a dynamic mode to recognize the gait, the hardware is simple to realize, but the technology is advanced, the gait recognition accuracy is high, the application prospect is wide, and the gait recognition method based on the image can be used for monitoring and analyzing the regional human health and can also be used for families to achieve customized service.
Description
Technical Field
The invention relates to the field of deep learning and human health, in particular to a CNN-LSTM-based human health deep learning prediction method.
Background
With the development of economy, the living standard of people is greatly improved, and meanwhile, the people pay more and more attention to health; the health condition of the human body is monitored, and the possible diseases of the human body are forecasted in advance, so that the method has important significance in the field of disease prevention and control. Studies have shown that human walking gait is closely related to its health status, and the common pathological gaits at present are scissors, mop, intoxicated, hemiplegic, paraplegic, gluteus maximus, gluteus medius, quadriceps, transregional, orthoarticular, teeter, flustered, dismay, crunchy, painful, etc., while the asynchronous state corresponds to different pathologies, such as: scissors step, two feet are separated from each other front and back, the back foot is stepped forward, the front foot is stepped forward to keep the distance between the front foot and the back foot when the two feet are separated, and the two feet are mostly in patients with cerebrovascular diseases, cerebral infarction, cerebral hemorrhage or spinal cord injury; the mop step, one leg of the patient is normal when walking, but the other leg is slowly dragged over due to the dysfunction of muscles or nerves; when a person is drunk, the gait is often out of line and shakes, and people with the gait have high suspicion of brain tumor, cerebral hemorrhage, cerebellar lesion and the like.
The slow change of the walking gait of the human body is considered as a normal retrogressive change and a normal category. If some gait abnormalities rapidly appear in a short time and the development speed is very high, and other nervous system symptoms accompany, a problem of one system is highly suspected. Based on the above, the invention provides a CNN-LSTM-based human health deep learning prediction method. Convolutional Neural Networks (CNN) are important algorithms in the field of deep learning, and are mainly used for feature extraction or classification of images, and a two-dimensional or higher-dimensional digital image can be directly used as an input of the Convolutional Neural Networks. The CNN algorithm has excellent feature extraction capability, and in the world ImageNet tournament of 2015, the image recognition error rate of the deep learning model based on the CNN algorithm is as low as 4.94%, which is lower than 5.1% of that of human beings. The Long Short-Term Memory network (LSTM) is mainly used for processing time series data and is widely applied to the fields of speech recognition, text recognition and the like. The invention provides a CNN-LSTM-based human health deep learning prediction method, which is characterized in that a lightweight and rapid multiplexing CNN model with strong feature mapping capability is constructed for pedestrian identification and gait feature extraction in an image, extracted gait features are input into an LSTM network, and then the LSTM identifies the gait features in a dynamic mode based on time sequence, judges the gait types of pedestrians and predicts the health condition of the pedestrians.
Disclosure of Invention
The invention aims to provide a CNN-LSTM-based human health deep learning prediction method; the invention obtains the image sequence of walking gait of human body through the camera, then uses CNN to extract the gait feature in each image, and uses the gait feature as the input of LSTM, and the LSTM algorithm pre-judges the health state of human body according to the change of the gait feature in the image sequence, and makes advance forecast. Its main advantage has:
1. the front-edge deep learning model is used for human health prediction, so that pedestrians in a video can be identified, and multi-person gaits can be detected simultaneously; the system has simple hardware equipment and wide application prospect, can be used for analyzing the health state of people in a specific environment and can also be used for families to achieve customized service;
2. the deep learning model designed by the invention is light in weight, has strong feature mapping capability, is high in speed for extracting image features, and can realize real-time discrimination of walking gaits of multiple people in a video;
the invention adopts the following technical scheme for realizing the purpose of the invention:
a CNN-LSTM-based human health deep learning prediction method mainly comprises two parts, namely hardware and software, wherein the hardware is a camera and a calculation module, and the software mainly comprises three modules, namely an image pedestrian detection module, a CNN-based feature extraction module and an LSTM-based feature sequence identification module. Firstly, acquiring a walking video of a user by using a camera, preprocessing each image in the video (adjusting the size, selecting a clear image, and the like), detecting each pedestrian in the image, extracting each pedestrian, taking the extracted pedestrian image as the input of a CNN (convolutional neural network) model, acquiring the gait feature of the pedestrian, and finally taking a gait feature sequence as the input of an LSTM (least Square, TM) to judge the health state of the human body. For a customized user, the user can be identified in the stage of acquiring the gait characteristics by the CNN, the health state database is established by combining the user information with the gait characteristics to realize long-term detection, more accurate health condition diagnosis is provided by analyzing the change condition of the gait of the user, and the health prediction of the user can be realized.
Preferably, the CNN-LSTM-based deep learning and prediction method for human health, provided by the invention, is used for judging the health condition of a user by combining an advanced deep learning model with a camera and analyzing the gait of the user when walking.
Preferably, the CNN-LSTM-based human health deep learning prediction method provided by the invention is used for constructing a light-weight CNN model with strong feature mapping capability for extracting pedestrian gait features in an image, inputting the extracted features into the LSTM, and identifying the pedestrian gait in a dynamic mode.
Preferably, according to the CNN-LSTM-based human health deep learning prediction method provided by the invention, each pedestrian image in a video is acquired through a pedestrian detection technology, and is sequentially input into a CNN feature extraction model, so that simultaneous diagnosis of multiple persons in the video is realized.
Preferably, the CNN-LSTM-based deep learning and prediction method for human health can realize customized service for family users, identify the users while extracting features, establish a database by combining user information with the gait of the users, analyze the gait features and the gait change conditions of the users, and provide more accurate health diagnosis and prediction.
Preferably, the CNN-LSTM-based human health deep learning prediction method provided by the invention has the advantages that fast-RCNN technology is adopted for pedestrian detection in the video, so that rapid detection of pedestrians is realized; before pedestrian detection, images in the video are preprocessed, and blurred images are discarded, so that the recognition rate is improved.
Preferably, in order to realize the real-time extraction of pedestrian gait features in a video and have strong feature mapping capability, the deep learning and prediction method for human health based on the CNN-LSTM provided by the invention constructs a lightweight CNN model containing 11 convolutional layers, the pixels of an input image are set to be 120 × 120, and a multi-step long convolutional layer is selected to replace a pooling layer for the maximum acceleration of feature extraction speed.
Has the advantages that:
compared with the prior art, the invention has the beneficial effects that: the advanced deep learning model is combined with the camera to predict the health state of the human body, the hardware is simple to implement, the technology is advanced, and the application prospect is wide. In the system, the CNN is combined with the LSTM to realize the dynamic identification of the gait of the pedestrian, and compared with the static gait feature identification based on a single image, the dynamic identification mode based on the sequence has higher identification rate; by means of the video pedestrian detection technology, real-time detection of multiple pedestrians in a video can be achieved, and by combining the big data technology, health assessment and prediction of people in a certain area can be achieved; for a home user, customized service can be realized, the user is identified while the gait features are extracted by the CNN, then a health state database is established by combining the user information with the gait features, the change condition of the gait of the user is analyzed through long-term observation, more accurate health condition diagnosis is provided, and meanwhile, the health condition of the user is predicted; the lightweight multiplexing CNN model has the advantages of strong feature mapping capability, high speed and capability of realizing real-time feature extraction.
Drawings
FIG. 1 System Structure
FIG. 2 pedestrian detection Process
FIG. 3 is a simplified structural diagram of a lightweight CNN model constructed by the present invention
Figure 4 gait recognition schematic diagram based on LSTM
Training convergence rate comparison chart of lightweight CNN constructed in FIG. 5
The specific implementation mode is as follows:
a human health deep learning prediction method based on CNN-LSTM, its implementation process is shown as figure 1, firstly use the camera to obtain the user walking video, secondly pre-process each image in the video (adjust size, select clear image, etc.), then detect each pedestrian in the image, and extract each pedestrian, then regard pedestrian's image extracted as the input of CNN model, and then obtain the gait characteristic of the pedestrian, finally regard the gait characteristic sequence formed by many pictures obtained as the input of LSTM, differentiate its walking gait type, thus predict the human health state; for customized users, the users can be identified in the stage of CNN acquiring gait characteristics, the health state database is established by combining the user information with the gait characteristics, long-term detection is realized, and more effective health condition prediction is provided.
The invention adopts a dynamic mode for identifying the gait of the pedestrian, namely, a camera is used for acquiring the gait video of the detected person when the detected person walks, and the walking gait of the detected person is judged by analyzing a series of walking images of the detected person so as to analyze the health state of the detected person. In order to guarantee the gait recognition rate, firstly, images acquired by a camera are screened to remove blurred pictures, then the size of the images is adjusted, then the image pedestrian detection is carried out to find out each pedestrian in the images, and then the gait of each pedestrian is judged. Because the invention judges the health state based on the gait, and the walking speed of the person is relatively slow, 5 video images are selected from the video images obtained by shooting every second in order to reduce the calculation amount of hardware. The invention is further illustrated by the following specific examples.
1. Image processing and screening
For image screening, the image is generally preprocessed, then the definition is judged, and finally a clear image is selected. On the premise of ensuring that a high-quality image is selected, in order to reduce the calculated amount as much as possible, the image preprocessing operation of the invention comprises image size adjustment, gaussian denoising and graying.
The definition of the image is generally expressed by gradient, and the larger the gradient is, the sharper the edge of the image is, and otherwise, the image is blurred. The method selects the Tenengrad function to judge the definition of the image. The Tenengrad function uses a Sobel operator to extract gradient values in the horizontal direction and the vertical direction, and the larger the value of the average gray value of the image processed by the Sobel operator is, the clearer the image is represented. Let G x And G y For the Sobel convolution kernel, I (x, y) is a pixel point, and the calculation of the Tenengrad gradient function is shown as the following formula:
2. pedestrian detection
With the rapid development of the depth science, the image detection technology is more and more mature, and at present, models for image detection are RCNN, fast-RCNN and the like, wherein the network structure of the Fast-RCNN is the most complex, but the image detection speed and the detection accuracy rate are also the highest. The implementation of fast-RCNN is mainly divided into three processes:
first, an input picture is processed by a pre-training CNN model to obtain a convolution feature map (conv feature map).
The extracted convolution feature map is then processed by the RPN (Region pro-position Network) which is used to find a predefined number of regions (bounding boxes) that may contain the target. The method comprises two parts, namely coordinates x and y and width and height w and h corresponding to a central anchor point of a predicted target area; and the second is used for judging whether the area is a foreground or a background.
And finally, classifying the content in the boundary frame based on the R-CNN module, and adjusting the coordinates of the boundary frame.
In order to ensure the real-time performance of the system, the invention uses the fast-RCNN model for pedestrian detection in the image, improves the traditional fast-RCNN model, and uses the constructed multiplexing CNN for feature extraction in the first step, and the implementation process is shown in FIG. 3.
3. CNN feature extraction
The CNN has excellent performance on image feature extraction or recognition, and researches show that the deeper the network of the CNN is, the stronger the feature learning capability is, but as the network is deepened, the network becomes more and more difficult to train, and the calculated amount of the model becomes larger and larger. In the convolutional neural network, a convolutional layer is used for feature extraction, the larger the convolutional kernel is, the more local related information is extracted, a pooling layer is mainly used for reducing the dimension of a feature map and is insensitive to the distortion and rotation of an image, and the essence of batch normalization is data normalization. And (3) constructing the CNN model, namely selecting different layers to construct a deep network according to actual requirements. In the invention, the CNN is used for extracting gait features in pictures and identifying pedestrians, and a lightweight composite CNN model with strong feature mapping capability is constructed, so that the identification of pedestrians and gait features can be realized, and the method is specifically shown in figure 3.
In FIG. 3, 'ker' represents the convolution kernel size, 'str' is the convolution step size, 'B _ Norm' is the batch normalization layer. In order to realize the lightweight of the model, the constructed CNN model selects 11 convolution layers and 6 batch normalization layers. The images input into the CNN are preprocessed and screened, and the problems of rotation and distortion do not exist, so that the multi-step long convolution layer is used for replacing the pooling layer, and the calculated amount of the model is reduced. In order to improve the convergence rate of the model, the multiplexing-type CNN model constructed by the invention adopts local residual connection, and a batch normalization layer is connected behind each residual layer, as shown in FIG. 3. To reduce the computation of LSTM while achieving more efficient gait characteristics, the second fully connected layer of the CNN model is set to 32.
4. LSTM-based gait recognition
LSTM is a more effective technique for solving the problem of long order dependence and is widely used in natural language processing and speech recognition. The core of LSTM improvement is to add a cell that determines whether information is remembered, and in the LSTM algorithm, each cell is connected with three gates, namely a forgetting gate, an input gate and an output gate, as shown in fig. 4. f. g and h are activation functions. In the LSTM information transmission process, the output of Input Gate is as follows:
in formula 2, x t For Input of Input Gate at the present moment, s t-1 For the outputs of all cells in the block at the previous time, b t-1 The outputs of other blocks at the last moment. The output of Forget Gate is as follows:
the Output of Output Gate is as follows:
the output of the Cell is as follows:
in the present invention, the output of the gait feature acquired by the CNN feature extraction model is set to 32, so the input width of the LSTM is set to 32.
Experimental verification section
In order to verify the convergence rate of the multiplexing CNN model constructed by the invention, the multiplexing CNN model is used for classification of a cifar-10 public data set, the resolution of a picture of the cifar-10 is 32 x 32, the step length of a convolution kernel of 5 is set as 1, and the rest is unchanged; in addition, a well-known deep learning model Vgg with 11 convolutional layers was also used for classification of the cifar-10 data, and the results were compared. Since the input to the Vgg model is 224 × 224, 5 pooling layers are included; to accommodate the size of the cifar-10 image data, vgg-11 was pooled first two layers and the convolution kernel of the first convolution layer was set to 3 x 3. The convergence rate of the two models in training is shown in fig. 5.
In FIG. 5, 'Vgg-11' is a training error curve of a classic network Vgg-11 on cifar-10, 'My-CNN' is an error curve of a lightweight CNN model constructed by the invention on cifar-10, and a comparison of the error curves of the two models shows that the lightweight CNN constructed by the invention has a faster error convergence rate. In conclusion, the lightweight CNN constructed by the invention not only has rapidity, but also has strong characteristic mapping capability.
It will be evident to those skilled in the art that the invention is not limited to the details of the foregoing illustrative embodiments, and that the present invention may be embodied in other specific forms without departing from the spirit or essential attributes thereof. The present embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the invention being indicated by the appended claims rather than by the foregoing description, and all changes which come within the meaning and range of equivalency of the claims are therefore intended to be embraced therein. Any reference sign in a claim should not be construed as limiting the claim concerned.
Furthermore, it should be understood that although the present specification describes embodiments, not every embodiment includes only a single embodiment, and such description is for clarity purposes only, and it is to be understood that all embodiments may be combined as appropriate by one of ordinary skill in the art to form other embodiments as will be apparent to those of skill in the art from the description herein.
Claims (7)
1. A CNN-LSTM-based human health deep learning prediction method is characterized by comprising the following steps: the system mainly comprises two parts, namely hardware and software, wherein the hardware is a camera and a calculation module, and the software mainly comprises three modules, namely a pedestrian detection module in an image, a CNN-based feature extraction module and an LSTM-based feature sequence identification module; firstly, acquiring a walking video of a user by using a camera, preprocessing each image in the video (adjusting the size, selecting a clear image, and the like), detecting each pedestrian in the image, extracting each pedestrian, taking the extracted pedestrian image as the input of a CNN (convolutional neural network) model to acquire the gait feature of the pedestrian, and finally taking a gait feature sequence as the input of an LSTM (least Square transition TM) to judge the health state of the human body; for a customized user, the user can be identified in the stage of acquiring the gait characteristics by the CNN, the health state database is established by combining the user information with the gait characteristics to realize long-term detection, more accurate health condition diagnosis is provided by analyzing the change condition of the gait of the user, and the health prediction of the user can be realized.
2. The CNN-LSTM-based deep learning prediction method for human health as claimed in claim 1, wherein: the advanced deep learning model is combined with a camera, and the health condition of the user is judged by analyzing the gait of the user when walking.
3. The CNN-LSTM-based deep learning prediction method for human health as claimed in claim 1, wherein: and constructing a light-weight CNN model with strong feature mapping capability for extracting pedestrian gait features in the image, inputting the extracted features into the LSTM, and judging the health condition of the user in a dynamic mode.
4. The CNN-LSTM-based deep learning prediction method for human health as claimed in claim 1, wherein: and acquiring each pedestrian image in the video through a pedestrian detection technology, and sequentially inputting the pedestrian image into the CNN feature extraction model so as to realize simultaneous diagnosis of multiple people in the video.
5. The CNN-LSTM-based deep learning prediction method for human health as claimed in claim 1, wherein: for the family user, the customized service can be realized, the user is identified while the characteristics are extracted, then the user information is combined with the gait to establish a database, the gait characteristics and the gait change condition of the user are analyzed, and more accurate health diagnosis and prediction are provided.
6. The CNN-LSTM-based deep learning prediction method for human health as claimed in claim 1, wherein: pedestrian detection in the video adopts the fast-RCNN technology to realize rapid detection of pedestrians; before pedestrian detection, images in the video are preprocessed, and blurred images are discarded, so that the recognition rate is improved.
7. The CNN-LSTM-based deep learning and prediction method for human health according to claim 1, characterized in that: in order to realize the real-time extraction of pedestrian gait features in a video and have strong feature mapping capability, a lightweight CNN model containing 11 convolutional layers is constructed, the pixels of an input image are set to be 120 x 120, and in order to accelerate the feature extraction speed to the maximum, a multi-step long convolutional layer is selected to replace a pooling 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 CN109902558A (en) | 2019-06-18 |
| CN109902558B true 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) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110390298B (en) * | 2019-07-23 | 2022-01-11 | 曲彦隆 | Gait simulation prediction system and simulation prediction method |
| 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 |
| CN111639533A (en) * | 2020-04-28 | 2020-09-08 | 深圳壹账通智能科技有限公司 | Gait feature-based posture detection method, device, equipment 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 |
| 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 |
| 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 |
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 (3)
| 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》.2018,(第20期), * |
| 基于深度学习的图像识别算法研究;衣世东;《网络安全技术与应用》;20180115(第01期);全文 * |
| 基于深度学习的实时人体动作识别监控方法;刘安祺等;《黑龙江科学》;20181023(第20期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109902558A (en) | 2019-06-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109902558B (en) | CNN-LSTM-based human health deep learning prediction method | |
| CN106778687B (en) | Fixation point detection method based on local evaluation and global optimization | |
| CN110464366A (en) | A kind of Emotion identification method, system and storage medium | |
| CN103699904B (en) | The image computer auxiliary judgment method of multisequencing nuclear magnetic resonance image | |
| CN110084149B (en) | Face verification method based on hard sample quadruple dynamic boundary loss function | |
| JP2018055470A (en) | Facial expression recognition method, facial expression recognition apparatus, computer program, and advertisement management system | |
| CN110929687B (en) | Multi-user behavior recognition system based on key point detection and working method | |
| CN109903339B (en) | Video group figure positioning detection method based on multi-dimensional fusion features | |
| CN107463917A (en) | A kind of face feature extraction method merged based on improved LTP with the two-way PCA of two dimension | |
| CN111210415B (en) | Method for detecting facial expression hypo of Parkinson patient | |
| CN110188715A (en) | A kind of video human face biopsy method of multi frame detection ballot | |
| CN112560810A (en) | Micro-expression recognition method based on multi-scale space-time characteristic neural network | |
| CN114359637A (en) | Brain medical image classification method and device | |
| Nebout et al. | Predicting saliency maps for ASD people | |
| CN115024706A (en) | Non-contact heart rate measurement method integrating ConvLSTM and CBAM attention mechanism | |
| CN112766165A (en) | Falling pre-judging method based on deep neural network and panoramic segmentation | |
| CN106529441A (en) | Fuzzy boundary fragmentation-based depth motion map human body action recognition method | |
| CN107862298A (en) | It is a kind of based on the biopsy method blinked under infrared eye | |
| Sakthimohan et al. | Detection and Recognition of Face Using Deep Learning | |
| CN111881818B (en) | Medical action fine-grained recognition device and computer-readable storage medium | |
| CN112488165A (en) | Infrared pedestrian identification method and system based on deep learning model | |
| CN109165542A (en) | Based on the pedestrian detection method for simplifying convolutional neural networks | |
| CN107707975A (en) | Video intelligent clipping method based on monitor supervision platform | |
| CN113762009A (en) | Crowd counting method based on multi-scale feature fusion and double-attention machine mechanism | |
| Li et al. | Dilated spatial–temporal convolutional auto-encoders for human fall detection in surveillance videos |
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 |