CN113143223A

CN113143223A - Edge artificial intelligence infant monitoring method

Info

Publication number: CN113143223A
Application number: CN202110006391.8A
Authority: CN
Inventors: 崔炜; 韩福鑫; 汪陈松; 李洋
Original assignee: Changchun University of Science and Technology
Current assignee: Changchun University of Science and Technology
Priority date: 2021-01-05
Filing date: 2021-01-05
Publication date: 2021-07-23

Abstract

The invention discloses a marginal artificial intelligence infant monitoring method, which is used for acquiring physiological information and posture information of infants through a millimeter wave radar and binocular vision and monitoring the physiological information and the posture information in a deep learning mode. The physiological information comprises heartbeat information and respiratory information, the heartbeat information and the respiratory information are firstly sent to a main network with characteristic extraction, the characteristic information of the physiological information is extracted, and then the heartbeat information and the respiratory information are monitored through a prediction network; the posture information comprises the posture and the bedding and clothing state of the infant, the improved darknet53 is selected as a main network for feature extraction, then the posture information of the infant is sent to the improved feature extraction network, the features of the posture information are extracted, and finally the posture and the bedding and clothing state of the infant are monitored through a prediction network. The full-day non-contact infant monitoring system realizes the full-day non-contact infant monitoring of monitoring the heartbeat and the breathing of the infant in real time and monitoring the posture and the bedding and clothing state of the infant at the same time.

Description

Edge artificial intelligence infant monitoring method

Technical Field

The invention relates to the field of infant monitoring, in particular to an edge artificial intelligence infant monitoring method, which realizes non-contact type infant monitoring.

Background

Infants before 3 years old do not have complete self-help ability, infants in the small months are completely without self-help ability, a guardian needs to spend a great deal of energy to nurse the infants to avoid accidents, but hundreds of mi are sparse, so that accidents are caused by carelessness of monitoring, and in order to make up for the problem, various infant monitoring methods are endless, however, the existing infant monitoring method generally only performs the crying detection and bedside distance detection of the infant, and cannot meet the actual requirements of infant monitoring, if the infant has a suffocation risk due to bad physical condition, the infant is often accompanied by body movement and turning over, the hidden danger of suffocation caused by the fact that the bedding is dropped and catching a cold or covers the mouth and nose exists, and the like, the guardian cannot timely detect the abnormality and can put the infant in danger, in order to remind a guardian of the abnormal posture of the infant, the posture detection of the infant is required; the infant mistakenly swallows the toy, swallows large-particle food and chokes milk, and suffocates, but the infant is suffocated and dies due to the fact that a guardian cannot timely perceive the infant when the infant suffocates, so that the heartbeat and the breath of the infant need to be monitored, and the guardian can master the physiological health condition of the infant in real time. However, at present, physiological health monitoring needs to be carried out by equipment, convenience is lacked, and infants wear the monitoring equipment for a long time to generate uncomfortable feeling and are not suitable for monitoring all day long, so that the edge artificial intelligence infant monitoring method is provided, and the non-contact monitoring of the infants all day long is realized through binocular vision and a millimeter wave radar. The infant sleeping posture monitoring system has the advantages that various information of infants is transmitted to the computer control end, sleeping postures, quilt kicking, breathing and heartbeat conditions of the infants are mastered in real time, potential risks are timely given out alarms, a guardian can timely handle the potential risks, and potential hidden dangers are eliminated.

Disclosure of Invention

The invention aims to provide a non-contact type infant monitoring method, which can monitor the heartbeat, the breathing, the posture and the bedding and clothing state of an infant all day long without causing discomfort to the infant;

the technical solution for realizing the invention is as follows: an edge artificial intelligence infant monitoring method comprises the following steps:

the method comprises the following steps: acquiring heartbeat signals and respiration signals of infants through a millimeter wave radar, acquiring current sleep posture information through binocular vision, and automatically switching the binocular vision according to brightness of acquired images to ensure that the acquired images are clear;

step two: selecting the darknet53 as a main feature extraction network, adding 4 residual error units on the basis of the darknet53 main feature extraction network, and continuously performing down-sampling once to obtain higher-layer semantic information and improve the richness of main network feature extraction;

step three: sending the acquired heartbeat signal and respiratory signal into a trained feature extraction backbone network, extracting feature information of a physiological signal, detecting respiratory information and heartbeat information through a prediction network, and sending a remote alarm signal to prompt a guardian by a control end in time when the prediction network detects that the physiological information is abnormal;

the specific treatment steps are as follows:

1) the method comprises the steps that a radar transmits linear frequency modulation continuous waves to an infant needing to measure vital signs; subsequently, processing the echo signal reflected by the human body to acquire an intermediate frequency signal; performing band-pass filtering on the intermediate frequency signal, acquiring a respiratory signal by adopting a 0.6-0.9 Hz band-pass filter, and acquiring a heartbeat signal by adopting a 1.3-2.1 Hz band-pass filter;

2) reducing the dimensions of the respiration signals and the heartbeat respiration to obtain two-dimensional characteristic data;

3) sending the two-dimensional feature data into an improved feature extraction network, and extracting feature information of the physiological signal by the feature extraction network;

4) and detecting the extracted characteristic information through a prediction network, and sending a remote alarm signal to prompt a guardian by a control end after detecting the abnormal physiological signal.

Step four: the acquired posture information is sent to an improved darknet53 characteristic extraction backbone network, the characteristics of the posture information are extracted, the posture and the bedding state of the infant are analyzed through a prediction network, and when the prediction network monitors that the posture information is abnormal, the control end actually sends a remote alarm signal to prompt a guardian.

The specific treatment steps are as follows:

1) switching a color camera and an infrared camera of binocular vision according to the brightness of the acquired image information to acquire clear image information;

2) sending the acquired image information into an improved feature extraction network to acquire the posture features of the infants;

3) and identifying the acquired posture characteristics through a prediction network, judging the posture and the bedding and clothing state of the infant, and sending remote alarm information to prompt a guardian in time by the control terminal after detecting the abnormal posture.

Has the advantages that: compared with the prior art, the technical scheme of the invention has the following beneficial technical effects:

the infant monitoring method realizes the non-contact type infant monitoring, and is an innovation of the infant monitoring method. The physical state and the physiological health of the infant are monitored, and the infant monitoring system is more practical than the traditional infant monitoring system;

in addition, the infant is monitored all day long by adopting a deep learning mode, and compared with the traditional infant monitoring method, the infant body posture monitoring method has the advantages that the body posture of the infant is monitored, and a guardian can timely master the posture and the bedding and clothing state of the infant;

in addition, the millimeter wave radar is adopted to carry out non-contact physiological health monitoring, and compared with a method for acquiring physiological information by a wearable device, the method greatly improves user experience.

In addition, a thought is provided for other infant monitoring methods, and various functions can be continuously improved subsequently, so that the infant monitoring method achieves the purpose of intellectualization.

Drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the embodiments or the prior art descriptions will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without inventive exercise.

FIG. 1 is a schematic flow chart of a method for monitoring an infant with edge artificial intelligence according to the present invention

FIG. 2 is a schematic view of a physiological information processing flow provided by the present invention

FIG. 3 is a schematic diagram of the processing flow of the posture information provided by the present invention

Detailed Description

In the following description, for purposes of explanation and not limitation, specific details are set forth, such as particular system structures, techniques, etc. in order to provide a thorough understanding of the embodiments of the invention. It will be apparent, however, to one skilled in the art that the present invention may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, and methods are omitted so as not to obscure the description of the present invention with unnecessary detail.

The invention is further illustrated by the following examples and figures of the specification.

Example 1

A marginal artificial intelligence infant monitoring method, as shown in fig. 1, comprising the steps of:

1) acquiring physiological information and posture information: carrying out non-contact acquisition on heartbeat and respiration information of the infant by using a millimeter wave radar; acquiring posture information by using binocular vision;

the radar is a frequency modulation continuous wave radar which transmits triangular waves by utilizing a linear frequency modulation technology; binocular vision refers to a color camera and an infrared camera;

2) information preprocessing: respectively preprocessing the physiological information and the posture information to enable the physiological information and the posture information to meet the input requirements of corresponding models in different modes;

3) the processed information is sent to a feature extraction network, and the features of the physiological information and the features of the posture information are respectively extracted;

4) the two extracted characteristic information are sent to a prediction network, the prediction network monitors the heartbeat and the breath of the infant, simultaneously monitors the posture and the quilt covering condition of the infant, and sends the final analysis result to a control end;

5) and after receiving the abnormal information, the control end sends a remote alarm signal to prompt the guardian in time.

Example 2

The physiological information processing flow, as shown in fig. 2, includes the following steps:

1) radar echo signal processing: processing echo signals reflected by a human body to acquire intermediate frequency signals; performing band-pass filtering on the intermediate frequency signal, acquiring a respiratory signal by adopting a 0.6-0.9 Hz band-pass filter, and acquiring a heartbeat signal by adopting a 1.3-2.1 Hz band-pass filter;

2) carrying out spectrum analysis on the acquired heartbeat signal and the acquired respiration to acquire frequency domain characteristics and time domain characteristics; carrying out ICA dimension reduction on the heartbeat characteristic and the respiration characteristic to obtain two-dimensional data;

3) sending the heart beat information and the breathing information with reduced dimension into an improved feature extraction network to extract the feature information of the physiological signal;

4) and sending the extracted characteristic information to a prediction network, analyzing the heartbeat breathing condition by the prediction network, transmitting the abnormal condition to a control end, and sending a remote alarm signal to prompt a guardian by the control end in time.

Example 3

The process of processing the attitude information, as shown in fig. 3, includes the following steps:

1) acquiring image information: the binocular vision acquires the posture information of the infant, and the system automatically switches the color camera and the infrared camera according to the illumination intensity of the environment, so that the definition of the acquired image is ensured;

2) decomposing the sleep condition acquired by the camera frame by frame, decomposing the video into image information of one frame and one frame, and sending the image into an improved feature extraction network to acquire feature information of the posture of the infant;

3) and sending the extracted characteristic information to a prediction network, identifying the posture and the bedding and clothing state of the infant by the prediction network, and sending a remote alarm signal to prompt a guardian in time by a control end for abnormal posture information.

Claims

1. An edge artificial intelligence infant monitoring method is characterized in that

Acquiring physiological information of infants, and monitoring abnormal heartbeat and respiration states; acquiring posture information for monitoring the posture and the state of the bedding and clothing;

the network structure is characterized in that: firstly, the darknet53 is selected as a main feature extraction network, 4 residual error units are added on the basis of the darknet53 main feature extraction network, down sampling is continuously carried out for one time, semantic information of a higher layer is obtained, and the richness of main network feature extraction is improved.

2. The method for monitoring the infant with the marginal artificial intelligence according to claim 1, wherein the monitoring of the abnormal heartbeat and respiration state comprises the following steps:

step 1: transmitting linear frequency modulation continuous waves to the infants needing to measure vital signs through a radar; subsequently, processing the echo signal reflected by the human body to acquire an intermediate frequency signal; performing band-pass filtering on the intermediate frequency signal to respectively obtain a heartbeat signal and a respiration signal;

step 2: sending the acquired heartbeat signal and the acquired respiratory signal into an improved feature extraction network, and extracting feature information of the physiological signal by the feature extraction network;

and step 3: and detecting the extracted characteristic information through a prediction network, and sending a remote alarm signal to prompt a guardian by a control end after detecting the abnormal physiological signal.

3. The method for monitoring infant with marginal artificial intelligence according to claim 1, wherein the posture and bedding condition monitoring comprises the following steps:

step 1: switching a color camera and an infrared camera of binocular vision according to the brightness of the acquired image information to acquire clear image information;

step 2: sending the acquired image information into an improved feature extraction network to acquire the posture features of the infants;

and step 3: and identifying the acquired posture characteristics through a prediction network, judging the posture and the bedding and clothing state of the infant, and sending remote alarm information to prompt a guardian when an abnormal posture is detected by the control terminal.