CN107609477A - It is a kind of that detection method is fallen down with what Intelligent bracelet was combined based on deep learning - Google Patents
It is a kind of that detection method is fallen down with what Intelligent bracelet was combined based on deep learning Download PDFInfo
- Publication number
- CN107609477A CN107609477A CN201710676771.6A CN201710676771A CN107609477A CN 107609477 A CN107609477 A CN 107609477A CN 201710676771 A CN201710676771 A CN 201710676771A CN 107609477 A CN107609477 A CN 107609477A
- Authority
- CN
- China
- Prior art keywords
- sample set
- training
- behavior
- belief network
- deep learning
- 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
Landscapes
- Image Analysis (AREA)
Abstract
Detection method is fallen down with what Intelligent bracelet was combined based on deep learning the invention discloses a kind of.This method is that gathered data progress foreground extraction establishes Sample Storehouse, Sample Storehouse is used to train depth belief network, the depth belief network Model Identification completed using training is fallen down or fallen down this two class behavior, whether then detects target blood pressure pulse normally so as to judging whether target falls down；Activity recognition method based on deep learning need not carry out engineer as conventional machines learning method to feature extracting method, it can be trained and learn on video data, obtain maximally effective characterizing method, finally by recognizer together with the connected applications of Intelligent bracelet hardware-efficient, higher detection accuracy is obtained.
Description
Technical Field
The invention relates to the technical field of health monitoring of old people, in particular to a tumble detection method based on combination of deep learning and an intelligent bracelet.
Background
At present, china has become the country with the most old people in the world and is one of the countries with the fastest population aging and development speed. Children and children are not in the side and the old are in the lonely spirit and do not have the care of others. The falling event of solitary old people in daily life directly threatens the life and health of the old people. Therefore, the method can detect the falling behavior of the old timely and accurately, and is a key for solving the problems.
The selection of a suitable fall detection algorithm is a core part of the detection system, but from the present situation, fall detection methods fall into two broad categories: 1) Based on the video, obtaining a moving target from the video, and detecting the tumble by using a traditional machine learning algorithm; 2) Based on the wearable device, data obtained by the sensor is analyzed for fall detection. For the traditional machine learning algorithm of the type 1), the defects are that the traditional machine learning algorithm has higher requirements on the environment of a video or shooting conditions and the like, and the feature extraction algorithm is designed in a manual prior mode, so that the workload is large, particularly the two conditions of falling and lying are distinguished, the distinguishing difficulty is increased, and the recognition rate is low. With the wearable device of the type 2), the blood pressure and pulse data obtained by the sensor are not necessarily caused by a fall in case of abnormality.
Disclosure of Invention
In order to solve the above problems, the present invention provides a fall detection method based on a combination of deep learning and a smart band, which combines a recognition algorithm with sensor hardware in an efficient manner, so as to more accurately detect a fall event.
The technical scheme adopted by the invention for solving the problems is as follows:
a tumbling detection method based on combination of deep learning and a smart bracelet comprises the following steps:
acquiring human body image data of a known behavior, wherein the human body image data of the known behavior is the human body image data of a known fall or lying down;
establishing a known behavior sample set for the acquired human body image data with the known behavior, wherein the known behavior sample set comprises a training sample set and a testing sample set;
carrying out foreground extraction on the known behavior sample set to obtain a foreground image of the known behavior sample set;
establishing a deep belief network, training the deep belief network by using a foreground image of a training sample set in a known behavior sample set, and acquiring better parameters of the deep belief network;
inputting a foreground image of a test sample set in a known behavior sample set into a deep belief network for recognition test, acquiring optimal parameters, and completing a deep belief network model;
acquiring unknown behavior human body image data including a human body target, and establishing an unknown behavior sample set;
carrying out foreground extraction on the unknown behavior sample set to obtain a foreground image of the unknown behavior sample set;
identifying whether the unknown behavior is a fall or lying down through a deep belief network model;
if the unknown behavior is recognized to be a fall or lie down, whether the human target blood pressure pulse is abnormal or not is detected through the intelligent bracelet, and if the detected human target blood pressure pulse is abnormal, the fall is judged.
Further, the image data of the human body with the known behavior and the image data of the human body with the unknown behavior including the human body target are collected by the camera.
Further, a foreground map of the known behavior sample set or the unknown behavior sample set is obtained by performing foreground extraction on the known behavior sample set or the unknown behavior sample set by using background subtraction based on a Gaussian mixture model.
Further, the specific process of the background subtraction based on the gaussian mixture model is as follows:
for any pixel point time-varying sequence { X in image 1 ,X 2 ,…,X t And modeling based on a Gaussian mixture model, wherein the probability of the pixel value of the current observation point is as follows:
where k is the number of Gaussian models, ω i,t Is the weight of the ith Gaussian model at the moment t, and meets the requirementsμ i,t And sigma i,t The mean value and the variance of the ith Gaussian model at the moment t are respectively; where η is the gaussian probability density function:
get theThe following method is selected to update the weight, the mean value and the variance of the Gaussian mixture model:
ω n,t ＝(1-α)ω n,t-1 +αΜ n,t
μ n,t ＝(1-ρ n,t )μ n,t-1 +ρ n,t X t
in the formula, α is a learning rate and represents a background update rate; m is a group of n,t The model matching operator is a value of 1 when a new pixel matches and 0 otherwise.
Further, a layer-by-layer greedy training algorithm is adopted to train the deep belief network, the training process comprises pre-training, and the pre-training comprises the following steps:
taking a foreground graph in a training sample as input data, and starting unsupervised learning from bottom to top; the deep belief network consists of a plurality of layers of restricted Boltzmann machines, and the training process is that a first restricted Boltzmann machine is fully trained; fixing the weight and offset of the first restricted Boltzmann machine, and using the state of the recessive neuron as the input of the second restricted Boltzmann machine; then after the second limited Boltzmann machine is trained fully, stacking the second limited Boltzmann machine above the first limited Boltzmann machine; and repeating the steps until all the restricted Boltzmann machines are trained, and obtaining the parameters of each layer.
Further, the training process further comprises tuning, wherein the tuning comprises the following steps:
starting top-down supervised learning, namely training through data with labels, transmitting errors from top to bottom, and finely adjusting the network; the fine-tuning process comprises a cognitive process and a generating process, and the weights of other layers of limited Boltzmann machines are divided into upward cognitive weights and downward generating weights except for the top layer of limited Boltzmann machines; during the cognition process, abstract representation of each layer is generated through external features and cognition weights, and gradient descent is used for modifying the generation weights among the layers; and in the generation process, the state of the bottom layer is generated through the representation and the weight generation of the top layer, and the cognitive weight between the layers is modified, so that the better parameters of the deep belief network are obtained.
Further, the specific process of the identification test is as follows:
during behavior recognition testing, a cross validation method is adopted, namely acquired human body image data of 2 or more persons with known behaviors are divided into a training sample set and a testing sample set, wherein the data of 1 person is used as the testing sample set, and the data of the rest persons are used as the training sample set to train the deep belief network model; repeating the steps until the data of each person is used as a test sample set to obtain two classification results of falling or lying; and adjusting the parameters of the deep learning network according to the obtained result, obtaining the optimal parameters and finishing the deep belief network model.
And further, inputting the foreground image of the unknown behavior sample set into the deep belief network model to obtain corresponding output, and judging whether the unknown behavior is a fall or lies down according to the output.
Further, after the unknown behavior is recognized to fall down or lie down, the blood pressure and the pulse data of the target are collected by the intelligent bracelet and serve as detection values, and whether the target falls down or not is normally judged according to the detection values.
Further, the blood pressure and pulse data of the target are collected through a sensor of the intelligent bracelet; whether the detection value is normal or not is judged through a safety threshold value of blood pressure and pulse preset in a central processing unit, wherein indexes of the blood pressure threshold value comprise systolic pressure and diastolic pressure, the safe range of the systolic pressure is 90-120, and the safe range of the diastolic pressure is 60-90; the safety range of the pulse threshold is 60-100 times/minute.
The beneficial effects of the invention are: according to the tumble detection method based on the combination of the deep learning and the smart bracelet, the behavior recognition method based on the deep learning does not need to design a feature extraction method manually like a traditional machine learning method, training and learning can be carried out on image data, and a most effective characterization method is obtained; in addition, under the condition that the misjudgment rate of the two behaviors of falling and lying is high by the existing algorithm, the invention adds the detection of blood pressure and pulse data on the basis of the identification algorithm, and the two behaviors are very easy to judge due to the great difference of the numerical values of the blood pressure and the pulse when falling and lying.
Drawings
The invention is further illustrated with reference to the following figures and examples.
Fig. 1 is a flow chart of a fall detection method based on deep learning combined with a smart bracelet according to the present invention;
FIG. 2 is a schematic structural diagram of a restricted Boltzmann machine model;
fig. 3 is a schematic structural diagram of a deep belief network model.
Detailed Description
Referring to fig. 1, the fall detection method based on the combination of deep learning and smart band of the present invention includes the following steps:
step S1: the method comprises the steps that a camera collects human body image data of known behaviors;
wherein the human body image data of the known behavior is human body image data of a known fall or lying down.
Step S2: establishing a known behavior sample set;
specifically, a known behavior sample set is established for the collected human body image data with known behaviors; the known behavior sample set is divided into a training sample set and a testing sample set.
And step S3: carrying out foreground extraction on a known behavior sample set;
specifically, foreground extraction is carried out on a known behavior sample set to obtain a foreground image of the known behavior sample set; wherein, the background subtraction based on the Gaussian mixture model is used for foreground extraction, the specific process is as follows,
for any pixel point time-varying sequence { X in image 1 ,X 2 ,…,X t And modeling based on a Gaussian mixture model, wherein the probability of the pixel value of the current observation point is as follows:
wherein k is the number of Gaussian models, and is usually 3 to 5; omega i,t The weight of the ith Gaussian model at the moment t is satisfiedμ i,t And sigma i,t The mean value and the variance of the ith Gaussian model at the moment t are respectively; where η is the gaussian probability density function:
taking into account the complexity of the calculationIn practical application, the background usually changes, so that each gaussian distribution parameter in the mixture model needs to be updated, and the weight, the mean and the variance of the gaussian mixture model are updated by the following method:
ω n,t ＝(1-α)ω n,t-1 +αΜ n,t
μ n,t ＝(1-ρ n,t )μ n,t-1 +ρ n,t X t
in the formula, α represents a learning rate and represents a background update rate; m n,t The model matching operator is a value of 1 when a new pixel matches and 0 otherwise.
And step S4: training a deep belief network;
specifically, a deep belief network is established, the deep belief network is trained by using a foreground image of a training sample set in a known behavior sample set, and better parameters of the deep belief network are obtained; and training and learning the deep belief network by adopting a layer-by-layer greedy training algorithm.
Step S5: behavior recognition testing is carried out to obtain the optimal parameters of the deep belief network model;
specifically, a foreground image of a test sample set in a known behavior sample set is input into a deep belief network model obtained through training of a training sample set for behavior recognition test, a corresponding output result is obtained, and the deep belief network model is modified according to the output result to obtain an optimal parameter; during behavior recognition testing, a cross validation method is adopted, namely collected human body image data of 2 or more persons with known behaviors are divided into a training sample set and a testing sample set, wherein data of 1 person is used as the testing sample set, and data of the rest persons are used as the training sample set to train a deep belief network model; and repeating the steps until the data of each person is used as a test sample set to acquire the classification results of falling or lying down.
Step S6: the method comprises the following steps that a camera collects human body image data of unknown behaviors;
specifically, the unknown-behavior human body image data is human body actual behavior image data.
Step S7: establishing an unknown behavior sample set;
specifically, an unknown behavior sample set is established for the unknown behavior human body image data collected by the camera.
Step S8: carrying out foreground extraction on the unknown behavior sample set;
specifically, foreground extraction is performed on the unknown behavior sample set by adopting a background subtraction method based on a Gaussian mixture model, so that a foreground image of the unknown behavior sample set is obtained.
Step S9: judging whether the patient falls down or lies down;
specifically, a foreground image of the unknown behavior sample set is input into the completed deep belief network model, an identification result of the unknown behavior is obtained, whether the identified unknown behavior belongs to the two behaviors of falling or lying down is judged, if the identification result belongs to one of the two behaviors of falling or lying down, the step S10 is executed, and if not, the step S6 is returned to.
Step S10: the intelligent bracelet collects blood pressure and pulse data;
specifically, after the unknown behavior is recognized to fall down or lie down, the intelligent bracelet collects the blood pressure and the pulse data of the target as detection values, and whether the target falls down or not is normally judged according to the detection values.
Step S11: judging whether the detection value exceeds a safety threshold value;
specifically, blood pressure and pulse data are collected through a sensor of the intelligent bracelet; whether the detection value is normal is judged through a safety threshold value of blood pressure and pulse preset in a central processing unit, wherein indexes of the blood pressure threshold value comprise systolic pressure and diastolic pressure, the safe range of the systolic pressure is 90-120, and the safe range of the diastolic pressure is 60-90; the safety range of the pulse threshold is 60-100 times/minute; if the detected value is beyond the safe range, executing step S12, otherwise returning to step S6.
And step S12, sending out an alarm signal.
Referring to fig. 2, a component element of the deep belief network, the restricted boltzmann model, has two layers of neurons, the lower layer is called a visual layer, and is used for inputting training data; the upper layer is called hidden layer and is used as a characteristic detector; each layer can be represented by a vector, and each dimension, namely each small circle in the graph, represents a neuron; the intralayer variables are not connected, and the interlaminar variables are fully connected; the joint configuration between the visible layer and the hidden layer of the constrained boltzmann machine can be expressed as an energy function,
θ＝{W,a,b}
wherein v is i ,h j Respectively representing the states of the visible layer node and the hidden layer node, and generally taking 0 or 1; a is j ,b i Denotes an offset, W ij Represents the connection weight between them; the joint probability distribution of the configuration can be determined by boltzmann distribution:
the conditional probability between the visual and hidden layers is calculated as follows:
in formula, σ (x) = (1 + e) -x ) -1 The method is a nonlinear function of a neuron, and can obtain an updated value of a weight parameter of a restricted Boltzmann machine model by calculating a partial derivative of an approximate number:
in the formula, ε represents the learning rate,<·> data in order to input the training data, it is,<·> model the expected value after model training is obtained; generally, a sampling approximation of reconstructed data is adopted by a contrast divergence method to update the network weight; the input of the next layer comes from the output of the previous layer, and so on, and the input of the lowest layer generally comes from the input original training data, which in this embodiment is the foreground map of the known behavior sample set or the foreground map of the unknown behavior sample set.
Referring to fig. 3, the deep belief network is composed of a plurality of layers of boltzmann machines, which can be regarded as a generative model or a discriminant model; the process of training the deep belief network by adopting a layer-by-layer greedy training algorithm comprises pre-training and fine tuning;
the pre-training specifically comprises the steps of training by adopting a bottom-up unsupervised greedy training algorithm, namely fully training a first limited Boltzmann machine, fixing the weight and offset of the first limited Boltzmann machine, using the state of a recessive neuron of the first limited Boltzmann machine as the input of a second limited Boltzmann machine, after fully training the second limited Boltzmann machine, stacking the second limited Boltzmann machine on the first limited Boltzmann machine, and repeating the steps until the whole network training is finished;
the fine tuning specifically comprises the steps of carrying out tuning by adopting a top-down supervision wake-sleep algorithm, namely generating a node state of each layer through external characteristics and upward weight in a wake stage, and modifying downlink weight between layers by using gradient descent; in a sleep stage, the state of the bottom layer is generated through the representation of the top layer and the downward weight, and the upward weight between the layers is modified to obtain the better parameters of the network.
The above description is only a preferred embodiment of the present invention, and the present invention is not limited to the above embodiment, and the present invention shall fall within the protection scope of the present invention as long as the technical effects of the present invention are achieved by the same means.
Claims (10)
1. A tumbling detection method based on combination of deep learning and a smart bracelet is characterized in that: the method comprises the following steps:
acquiring human body image data of a known behavior, wherein the human body image data of the known behavior is the human body image data of a known fall or lying down;
establishing a known behavior sample set for the acquired human body image data with the known behavior, wherein the known behavior sample set comprises a training sample set and a testing sample set;
carrying out foreground extraction on the known behavior sample set to obtain a foreground image of the known behavior sample set;
establishing a deep belief network, training the deep belief network by using a foreground image of a training sample set in a known behavior sample set, and acquiring better parameters of the deep belief network;
inputting a foreground image of a test sample set in a known behavior sample set into a deep belief network for recognition test, acquiring optimal parameters, and completing a deep belief network model;
collecting unknown behavior human body image data including a human body target, and establishing an unknown behavior sample set;
carrying out foreground extraction on the unknown behavior sample set to obtain a foreground image of the unknown behavior sample set;
identifying whether the unknown behavior is a fall or lying down through a deep belief network model;
if the unknown behavior is recognized to be a fall or lie down, whether the human target blood pressure pulse is abnormal or not is detected through the intelligent bracelet, and if the detected human target blood pressure pulse is abnormal, the fall is judged.
2. The fall detection method based on deep learning and smart band combination according to claim 1, characterized in that: and acquiring the image data of the human body with the known behavior and the image data of the human body with the unknown behavior comprising the human body target through a camera.
3. The fall detection method based on deep learning and smart band combination according to claim 1, characterized in that: and obtaining a foreground image of the known behavior sample set or the unknown behavior sample set by performing foreground extraction on the known behavior sample set or the unknown behavior sample set by using background subtraction based on a Gaussian mixture model.
4. The fall detection method based on the combination of deep learning and smart band of claim 3, characterized in that: the specific process of the background subtraction based on the gaussian mixture model is as follows:
for any pixel point in the image time-varying sequence { X 1 ,X 2 ,…,X t And modeling based on a Gaussian mixture model, wherein the probability of the pixel value of the current observation point is as follows:
where k is the number of Gaussian models, ω i,t Is the weight of the ith Gaussian model at the moment t, and meets the requirementsμ i,t And sigma i,t The mean value and the variance of the ith Gaussian model at the moment t are respectively; where η is the gaussian probability density function:
getThe following method is selected to update the weight, the mean value and the variance of the Gaussian mixture model:
ω n,t ＝(1-α)ω n,t-1 +αΜ n,t
μ n,t ＝(1-ρ n,t )μ n,t-1 +ρ n,t X t
in the formula, α is a learning rate and represents a background update rate; m is a group of n,t For the model matching operator, the value is 1 when a new pixel is matched, otherwise it is 0.
5. The fall detection method based on the combination of deep learning and smart band of claim 1, characterized in that: training a deep belief network by adopting a layer-by-layer greedy training algorithm, wherein the training process comprises pre-training, and the pre-training comprises the following steps:
taking a foreground graph in a training sample as input data, and starting unsupervised learning training from bottom to top; the deep belief network consists of a plurality of layers of limited Boltzmann machines, and the training process is that a first limited Boltzmann machine is fully trained; fixing the weight and offset of the first restricted Boltzmann machine, and using the state of the recessive neuron as the input of the second restricted Boltzmann machine; then after the second limited Boltzmann machine is trained fully, stacking the second limited Boltzmann machine above the first limited Boltzmann machine; and repeating the steps until all the restricted Boltzmann machines are trained, and obtaining the parameters of each layer.
6. The fall detection method based on the combination of deep learning and smart band of claim 1 or 5, characterized in that: the training process further comprises tuning, wherein the tuning comprises the following steps:
starting top-down supervised learning, namely training through data with labels, transmitting errors from top to bottom, and finely adjusting the network; the fine adjustment process comprises a cognitive process and a generation process, and except for the top limited Boltzmann machine, the weights of other layers of limited Boltzmann machines are divided into an upward cognitive weight and a downward generation weight; during the cognition process, generating abstract representation of each layer through external features and cognition weights, and modifying the generation weights among the layers by using gradient descent; and in the generation process, the state of the bottom layer is generated through the representation and the weight generation of the top layer, and the cognitive weight between the layers is modified, so that the better parameters of the deep belief network are obtained.
7. The fall detection method based on the combination of deep learning and smart band of claim 1, characterized in that: the specific process of the identification test is as follows:
during behavior recognition testing, a cross validation method is adopted, namely acquired human body image data of 2 or more persons with known behaviors are divided into a training sample set and a testing sample set, wherein the data of 1 person is used as the testing sample set, and the data of the rest persons are used as the training sample set to train the deep belief network model; repeating the steps until the data of each person is used as a test sample set to obtain two classification results of falling or lying down; and adjusting the parameters of the deep learning network according to the obtained result, obtaining the optimal parameters and finishing the deep belief network model.
8. The fall detection method based on deep learning and smart band combination according to claim 1, characterized in that: and inputting the foreground image of the unknown behavior sample set into the deep belief network model to obtain corresponding output, and judging whether the unknown behavior is a fall or lies down according to the output.
9. The fall detection method based on the combination of deep learning and smart band of claim 1, characterized in that: after discerning that unknown action is fallen or lies down, whether normal judgement target of intelligent bracelet falls down according to the detected value as the detected value with the pulse data of target,.
10. The fall detection method based on the combination of deep learning and smart band of claim 9, characterized in that: the blood pressure and pulse data of the target are acquired through a sensor of the intelligent bracelet; whether the detection value is normal or not is judged through a safety threshold value of blood pressure and pulse preset in a central processing unit, wherein indexes of the blood pressure threshold value comprise systolic pressure and diastolic pressure, the safe range of the systolic pressure is 90-120, and the safe range of the diastolic pressure is 60-90; the safety range of the pulse threshold is 60-100 times/minute.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710676771.6A CN107609477A (en) | 2017-08-09 | 2017-08-09 | It is a kind of that detection method is fallen down with what Intelligent bracelet was combined based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710676771.6A CN107609477A (en) | 2017-08-09 | 2017-08-09 | It is a kind of that detection method is fallen down with what Intelligent bracelet was combined based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107609477A true CN107609477A (en) | 2018-01-19 |
Family
ID=61065009
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710676771.6A Pending CN107609477A (en) | 2017-08-09 | 2017-08-09 | It is a kind of that detection method is fallen down with what Intelligent bracelet was combined based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107609477A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108596150A (en) * | 2018-05-10 | 2018-09-28 | 南京大学 | A kind of Activity recognition system and its working method excluding abnormal operation |
| CN109213840A (en) * | 2018-09-12 | 2019-01-15 | 北京英视睿达科技有限公司 | Hot spot grid recognition methods based on multidimensional characteristic deep learning |
| CN109247914A (en) * | 2018-08-29 | 2019-01-22 | 百度在线网络技术（北京）有限公司 | Illness data capture method and device |
| CN109346166A (en) * | 2018-11-22 | 2019-02-15 | 深圳市云护宝计算机技术有限公司 | A kind of inpatient department intelligent medical care bracelet and its deep learning modeling method |
| WO2020019926A1 (en) * | 2018-07-27 | 2020-01-30 | 腾讯科技（深圳）有限公司 | Feature extraction model training method and apparatus, computer device, and computer readable storage medium |
| WO2020058763A1 (en) * | 2018-09-17 | 2020-03-26 | Vr Emoji Limited | Systems and methods for collecting data used in machine learning for object recognition |
| CN111626273A (en) * | 2020-07-29 | 2020-09-04 | 成都睿沿科技有限公司 | Fall behavior recognition system and method based on atomic action time sequence characteristics |
| CN113221661A (en) * | 2021-04-14 | 2021-08-06 | 浪潮天元通信信息系统有限公司 | Intelligent human body tumbling detection system and method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103793718A (en) * | 2013-12-11 | 2014-05-14 | 台州学院 | Deep study-based facial expression recognition method |
| US8898093B1 (en) * | 2012-06-25 | 2014-11-25 | The Boeing Company | Systems and methods for analyzing data using deep belief networks (DBN) and identifying a pattern in a graph |
| CN106097656A (en) * | 2016-08-22 | 2016-11-09 | 南京工程学院 | Old man care system based on Internet of Things |
| CN106548645A (en) * | 2016-11-03 | 2017-03-29 | 济南博图信息技术有限公司 | Vehicle route optimization method and system based on deep learning |
| CN106625714A (en) * | 2017-01-17 | 2017-05-10 | 五邑大学 | Monitoring robot for the old physical health condition detection |
-
2017
- 2017-08-09 CN CN201710676771.6A patent/CN107609477A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8898093B1 (en) * | 2012-06-25 | 2014-11-25 | The Boeing Company | Systems and methods for analyzing data using deep belief networks (DBN) and identifying a pattern in a graph |
| CN103793718A (en) * | 2013-12-11 | 2014-05-14 | 台州学院 | Deep study-based facial expression recognition method |
| CN106097656A (en) * | 2016-08-22 | 2016-11-09 | 南京工程学院 | Old man care system based on Internet of Things |
| CN106548645A (en) * | 2016-11-03 | 2017-03-29 | 济南博图信息技术有限公司 | Vehicle route optimization method and system based on deep learning |
| CN106625714A (en) * | 2017-01-17 | 2017-05-10 | 五邑大学 | Monitoring robot for the old physical health condition detection |
Non-Patent Citations (1)
| Title |
|---|
| 樊恒、徐俊等: "基于深度学习的人体行为识别", 《武汉大学学报·信息科学版》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108596150A (en) * | 2018-05-10 | 2018-09-28 | 南京大学 | A kind of Activity recognition system and its working method excluding abnormal operation |
| WO2020019926A1 (en) * | 2018-07-27 | 2020-01-30 | 腾讯科技（深圳）有限公司 | Feature extraction model training method and apparatus, computer device, and computer readable storage medium |
| US11538246B2 (en) | 2018-07-27 | 2022-12-27 | Tencent Technology (Shenzhen) Company Limited | Method and apparatus for training feature extraction model, computer device, and computer-readable storage medium |
| CN109247914A (en) * | 2018-08-29 | 2019-01-22 | 百度在线网络技术（北京）有限公司 | Illness data capture method and device |
| CN109213840A (en) * | 2018-09-12 | 2019-01-15 | 北京英视睿达科技有限公司 | Hot spot grid recognition methods based on multidimensional characteristic deep learning |
| CN109213840B (en) * | 2018-09-12 | 2023-05-30 | 北京英视睿达科技股份有限公司 | Hot spot grid identification method based on multidimensional feature deep learning |
| WO2020058763A1 (en) * | 2018-09-17 | 2020-03-26 | Vr Emoji Limited | Systems and methods for collecting data used in machine learning for object recognition |
| CN109346166A (en) * | 2018-11-22 | 2019-02-15 | 深圳市云护宝计算机技术有限公司 | A kind of inpatient department intelligent medical care bracelet and its deep learning modeling method |
| CN111626273A (en) * | 2020-07-29 | 2020-09-04 | 成都睿沿科技有限公司 | Fall behavior recognition system and method based on atomic action time sequence characteristics |
| CN113221661A (en) * | 2021-04-14 | 2021-08-06 | 浪潮天元通信信息系统有限公司 | Intelligent human body tumbling detection system and method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107609477A (en) | It is a kind of that detection method is fallen down with what Intelligent bracelet was combined based on deep learning | |
| CN102302370B (en) | Method and device for detecting tumbling | |
| CN109171707A (en) | A kind of intelligent cardiac figure classification method | |
| CN111436944B (en) | Falling detection method based on intelligent mobile terminal | |
| CN105139029B (en) | A kind of Activity recognition method and device of prison prisoner | |
| CN106580282A (en) | Human body health monitoring device, system and method | |
| CN108509897A (en) | A kind of human posture recognition method and system | |
| CN113012815B (en) | Multi-mode data-based parkinsonism health risk assessment method | |
| CN108958482B (en) | Similarity action recognition device and method based on convolutional neural network | |
| CN113516828B (en) | Drowning monitoring method, drowning monitoring device, drowning monitoring equipment and computer readable storage medium | |
| CN117598700B (en) | Intelligent blood oxygen saturation detection system and method | |
| Alharthi et al. | Deep learning for ground reaction force data analysis: Application to wide-area floor sensing | |
| CN117198468B (en) | Intervention scheme intelligent management system based on behavior recognition and data analysis | |
| CN108717548B (en) | Behavior recognition model updating method and system for dynamic increase of sensors | |
| CN110991471A (en) | Fault diagnosis method for high-speed train traction system | |
| Ning et al. | Fall detection algorithm based on gradient boosting decision tree | |
| Youn et al. | Wearable sensor-based biometric gait classification algorithm using WEKA | |
| CN110123484B (en) | Livestock delivery detection method and device | |
| Zheng et al. | Rapid screening of children with autism spectrum disorders through face image classification | |
| CN111646332A (en) | Method and system for identifying abnormal operation of elevator | |
| Nam et al. | Selective prediction with long short-term memory using unit-wise batch standardization for time series health data sets: algorithm development and validation | |
| CN115414054A (en) | Epilepsia electroencephalogram detection system based on feedforward pulse neural network | |
| CN114913585A (en) | Household old man falling detection method integrating facial expressions | |
| CN109359580A (en) | Footprint based on deep learning identifies and gait detection method and its device | |
| CN113095153A (en) | Mobile terminal human situation recognition method based on depth residual error network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180119 |
|
| RJ01 | Rejection of invention patent application after publication |