Children smart watch danger early warning method based on environment recognition

Abstract

The invention discloses a child smart watch danger early warning method based on environment recognition, and relates to the technical field of electronic products, and the method comprises the steps: installing a miniature camera on a smart watch, and debugging the camera; the camera captures a continuous image sequence in the environment in real time, and sequentially performs denoising, enhancement and graying processing on the captured image; extracting edge, texture and shape features from the preprocessed image; detecting the motion condition of an object in the environment; extracting the behavior state of the environment object and the behavior state of the child; training the model constructed by the random forest algorithm; and outputting the corresponding warning signal and the prompt signal to the child and the child guardian respectively. By setting the motion condition of the object in the detection environment, extracting the behavior state of the object in the environment and the behavior state of the child and obtaining the classification recognition model, accurate recognition of the child safety state is achieved, the child safety is improved, and accidents are reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of electronic products, and in particular to a danger warning method for a children's smart watch based on environment recognition. Background Art

[0002] In recent years, the smartwatch market has developed rapidly and has gradually become an indispensable part of modern life. With the improvement of people's health awareness and the advancement of technology, more and more consumers have begun to pay attention to the functionality of smartwatches, not only to display time, but also to their functions in health monitoring, communication, positioning, etc. Whether they are sports enthusiasts, students, or office workers, smartwatches meet people's needs for technology in daily life with their convenience and versatility.

[0003] However, in practical applications, how to monitor the status of children in real time and promptly detect and warn of potential safety hazards has become an urgent problem to be solved. Objects in the environment will constantly move and change, and environmental objects may also be blocked, all of which will affect the accuracy of recognition. Summary of the invention

[0004] In order to solve the above technical problems, a danger warning method for children's smart watches based on environment recognition is provided. This technical solution solves the problems raised in the above background technology.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is: A children's smart watch danger warning method based on environment recognition, comprising: Install a miniature camera on the smartwatch and debug the camera, adjusting its angle and focal length parameters to ensure that the camera captures a full view of the child's surroundings; The camera captures continuous image sequences in the environment in real time, and performs denoising, enhancement and grayscale processing on the captured images in turn; Extract edge, texture and shape features from the preprocessed image; Based on the image frame difference method, the motion of objects in the environment is detected by performing image dynamic detection on the continuous image sequence in the environment captured by the camera in real time; Through texture analysis of each feature, based on the movement of objects in the environment, the behavior state of environmental objects and the behavior state of children are extracted; Using the random forest algorithm, the behavioral states of environmental objects are linked to dangerous situations of children; The model constructed by the random forest algorithm is trained using a historical environmental monitoring data set containing various environmental scenarios to obtain a classification and recognition model; The continuous image sequence in the environment captured by the camera in real time is input into the classification and recognition model, and the corresponding warning signal and prompt signal are output to the child and the child's guardian respectively.

[0006] Preferably, the camera captures a continuous image sequence in the environment in real time, and performs denoising, enhancement and grayscale processing on the captured images in sequence, specifically including: Use image processing library to capture live images from camera; Use a median filter to remove noise from the image and replace the value of each pixel in the continuous image sequence of the environment with the median of its neighborhood pixel value; By analyzing the pixel value distribution of the image, the pixel value of each pixel is redistributed to even out the contrast of the image; Multiply all pixel values ​​of the image by a preset constant to enhance the brightness of the image; Obtain the weighted coefficients of human eye sensitivity to red, green and blue; Substitute the red, green and blue color components of each pixel in the continuous image sequence in the environment into the weighted average formula to obtain the grayscale value of each pixel in the continuous image sequence in the environment; The weighted average formula is: , In the formula, is the gray value of the pixel at the i-th row and j-th column, are the weighted coefficients of human eye sensitivity to red, green and blue, respectively. They are the red, green and blue color components of the pixel in the i-th row and j-th column respectively.

[0007] Preferably, the step of extracting edge, texture and shape features from the preprocessed image specifically includes: Get the third-order Laplace operator; Perform convolution operation on the third-order Laplace operator and the preprocessed image to obtain the new pixel value of each pixel; In the image after Laplace operation, the zero crossing points where the pixel value changes from positive to negative or from negative to positive are identified, and the zero crossing points are output as edge points of the image; Set the neighborhood of each pixel to 8 pixels within a 3×3 range centered on the pixel; Determine whether the grayscale value of each neighboring pixel is greater than or equal to the grayscale value of the central pixel. If so, output the binary bit of the neighboring pixel as 1; if not, output the binary bit of the neighboring pixel as 0; Combining 8 binary bits clockwise from the upper left corner, we get an 8-bit binary number. Output the decimal number corresponding to the binary number as the texture feature value of the pixel position; Calculate the frequency of occurrence of texture eigenvalues ​​in each image and construct a texture eigenvalue frequency histogram; Based on the figure surrounded by edge points, its area, perimeter and direction are output as shape features; The Laplace operator is: , In the formula, is the third-order Laplace operator; The convolution operation is: , In the formula, is the new pixel value of the pixel at the i-th row and j-th column.

[0008] Preferably, the image frame difference method is based on performing image dynamic detection on a continuous image sequence in the environment captured by a camera in real time to detect the motion of an object in the environment, specifically comprising: Continuously capture two adjacent frames of images from a continuous image sequence of the environment captured by the camera in real time; For each pixel position of the two frames of images, calculate the difference in grayscale values; The minimum amplitude of pixel value change that can be recognized by human eyes is used as the preset difference threshold; Determine whether the difference in the grayscale value of each pixel is greater than a preset difference threshold. If so, output that the pixel is in motion; if not, output that the pixel belongs to a static background; Cluster all pixels with motion to form connected regions, and output each connected region as a moving target in the environment; Repeatedly capture subsequent adjacent images to perform image dynamic detection, track the motion trajectory of each moving target and calculate the real-time motion speed of each moving target.

[0009] Preferably, the step of calculating the real-time motion speed of each moving target comprises the following steps: According to the moving trajectory of the moving target, the initial moving speed of the moving target is calculated; The minimum value of the initial movement speed is used as the child's movement speed; The child's movement speed is reversely superimposed on the initial movement speed to obtain the real-time movement speed of the moving target.

[0010] Preferably, the extracting the behavior state of the environmental objects and the behavior state of the children based on the movement of the objects in the environment through texture analysis of each feature specifically includes: Based on the texture analysis of each feature, locate the facial area of ​​the moving target; When the moving target does not have a facial region, the speed reduction when the distance between the moving target and the child is less than the safe distance is taken as the behavior state of the environmental object; When the moving target has a facial region, locating key feature points in the facial region of the moving target, wherein the key feature points include eyes, mouth and eyebrows; Analyze the texture of the facial area and extract facial expression features; Compare the extracted facial expression features with the known expression database to identify the current facial expression of the moving target; Count the number of blinks per unit time, and determine the opening and closing state of the eyes based on the occlusion of the pupils, and summarize and output the eye state of the moving target; Extract and classify features of images of the hand area to identify hand movements of moving targets; Track hand movements in continuous image frames, analyze the continuity and changes of hand movements, and output the gestures of moving targets; The current facial expression, eye state and gesture action of the moving target are regarded as the behavior state of the environmental object; The same analysis steps as the motion target are used to obtain the child’s current facial expression, eye state, and gesture movement, which are used as the child’s behavioral state.

[0011] Preferably, the training of the model constructed by the random forest algorithm using a historical environmental monitoring data set containing various environmental scenes to obtain a classification recognition model specifically includes: S101: Setting the historical environment monitoring database as a model training data set; S102: Randomly select a data subset that has not been selected from the historical environment monitoring database as the root node of the decision tree; S103: establishing a first leaf node at the root node according to different behavior states of the moving target as a splitting criterion; S104: establishing a second leaf node at the first leaf node based on the difference in the behavior status of the children as a splitting criterion; S105: Summarize the root node and all its leaf nodes and output them as a decision tree; S106: Determine whether corresponding decision trees are established for all data subsets in the historical environment monitoring database. If so, output all decision trees as classification and recognition models. If not, return to step S102.

[0012] Compared with the prior art, the present invention has the following beneficial effects: By setting up the detection of the movement of objects in the environment, extracting the behavioral status of environmental objects and the behavioral status of children and obtaining a classification and recognition model, real-time monitoring and early warning of the status of children and their guardians can be achieved, driving safety can be improved, and the occurrence of traffic accidents can be reduced. By preprocessing images (such as denoising, enhancement and grayscale processing) and using advanced classification and recognition algorithms (such as random forest algorithms), accurate identification of children's safety status can be achieved under complex lighting and occlusion conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a flow chart of the danger warning method of a children's smart watch based on environment recognition of the present invention; Figure 2 A flow chart of a method for sequentially performing denoising, enhancement and grayscale processing on a captured image according to the present invention; Figure 3 A flow chart of a method for extracting edge, texture and shape features from a preprocessed image according to the present invention; Figure 4 It is a flow chart of a method for performing image dynamic detection on a continuous image sequence in an environment captured in real time by a camera according to the present invention; Figure 5 A flow chart of a method for calculating the real-time motion speed of each moving target of the present invention; Figure 6 A flow chart of a method for extracting the behavior state of environmental objects and the behavior state of children according to the present invention; Figure 7 This is a flow chart of a method for training a model constructed using a random forest algorithm according to the present invention. DETAILED DESCRIPTION

[0014] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art may think of other obvious variations.

[0015] Reference Figure 1 As shown, a danger warning method for a children's smart watch based on environment recognition includes: Install a miniature camera on the smartwatch and debug the camera, adjusting its angle and focal length parameters to ensure that the camera captures a full view of the child's surroundings; The camera captures continuous image sequences in the environment in real time, and performs denoising, enhancement and grayscale processing on the captured images in turn; Extract edge, texture and shape features from the preprocessed image; Based on the image frame difference method, the motion of objects in the environment is detected by performing image dynamic detection on the continuous image sequence in the environment captured by the camera in real time; Through texture analysis of each feature, based on the movement of objects in the environment, the behavior state of environmental objects and the behavior state of children are extracted; Using the random forest algorithm, the behavioral states of environmental objects are linked to dangerous situations of children; The model constructed by the random forest algorithm is trained using a historical environmental monitoring data set containing various environmental scenarios to obtain a classification and recognition model; The continuous image sequence in the environment captured by the camera in real time is input into the classification and recognition model, and the corresponding warning signal and prompt signal are output to the child and the child's guardian respectively.

[0016] Reference Figure 2 As shown, the camera captures a continuous image sequence in the environment in real time, and performs denoising, enhancement and grayscale processing on the captured images in turn, specifically including: Use image processing library to capture live images from camera; Use a median filter to remove noise from the image and replace the value of each pixel in the continuous image sequence of the environment with the median of its neighborhood pixel value; By analyzing the pixel value distribution of the image, the pixel value of each pixel is redistributed to even out the contrast of the image; Multiply all pixel values ​​of the image by a preset constant to enhance the brightness of the image; Obtain the weighted coefficients of human eye sensitivity to red, green and blue; Substitute the red, green and blue color components of each pixel in the continuous image sequence in the environment into the weighted average formula to obtain the grayscale value of each pixel in the continuous image sequence in the environment; The weighted average formula is: , In the formula, is the gray value of the pixel at the i-th row and j-th column, are the weighted coefficients of human eye sensitivity to red, green and blue, respectively. They are the red, green and blue color components of the pixel in the i-th row and j-th column respectively.

[0017] Get the weighted coefficients of human eye sensitivity to red, green and blue. Usually, the human eye is most sensitive to green, second to red, and least sensitive to blue. Therefore, the weighted coefficients of green, red and blue can be set to 0.59, 0.30 and 0.11 respectively (these coefficients are based on the characteristics of the human visual system).

[0018] Reference Figure 3 As shown, the edge, texture and shape features extracted from the preprocessed image specifically include: Get the third-order Laplace operator; Perform convolution operation on the third-order Laplace operator and the preprocessed image to obtain the new pixel value of each pixel; In the image after Laplace operation, the zero crossing points where the pixel value changes from positive to negative or from negative to positive are identified, and the zero crossing points are output as edge points of the image; Set the neighborhood of each pixel to 8 pixels within a 3×3 range centered on the pixel; Determine whether the grayscale value of each neighboring pixel is greater than or equal to the grayscale value of the central pixel. If so, output the binary bit of the neighboring pixel as 1; if not, output the binary bit of the neighboring pixel as 0; Combining 8 binary bits clockwise from the upper left corner, we get an 8-bit binary number. Output the decimal number corresponding to the binary number as the texture feature value of the pixel position; Calculate the frequency of occurrence of texture eigenvalues ​​in each image and construct a texture eigenvalue frequency histogram; Based on the figure surrounded by edge points, its area, perimeter and direction are output as shape features; The Laplace operator is: , In the formula, is the third-order Laplace operator; The convolution operation is: , In the formula, is the new pixel value of the pixel at the i-th row and j-th column.

[0019] For each closed figure, its area, perimeter and direction are calculated as shape features. The area can be obtained by counting the number of pixels inside the figure; the perimeter can be obtained by traversing the edge points of the figure and calculating the sum of the distances between adjacent points; the direction can be obtained by calculating the angle between the long axis of the figure and the horizontal axis.

[0020] Reference Figure 4 As shown, based on the image frame difference method, the image dynamic detection is performed on the continuous image sequence in the environment captured by the camera in real time, and the motion of the objects in the environment is detected specifically including: Continuously capture two adjacent frames of images from a continuous image sequence of the environment captured by the camera in real time; For each pixel position of the two frames of images, calculate the difference in grayscale values; The minimum amplitude of pixel value change that can be recognized by human eyes is used as the preset difference threshold; Determine whether the difference in the grayscale value of each pixel is greater than a preset difference threshold. If so, output that the pixel is in motion; if not, output that the pixel belongs to a static background; Cluster all pixels with motion to form connected regions, and output each connected region as a moving target in the environment; Repeatedly capture subsequent adjacent images to perform image dynamic detection, track the motion trajectory of each moving target and calculate the real-time motion speed of each moving target.

[0021] Based on the sensitivity of the lighting changes in the environment, a difference threshold is preset. This threshold is used to distinguish the grayscale value changes caused by object motion and the grayscale value changes caused by lighting changes. The setting of the difference threshold needs to be adjusted according to the actual lighting conditions and camera performance.

[0022] Reference Figure 5 As shown, calculating the real-time motion speed of each moving target includes the following steps: According to the moving trajectory of the moving target, the initial moving speed of the moving target is calculated; The minimum value of the initial movement speed is used as the child's movement speed; The child's movement speed is reversely superimposed on the initial movement speed to obtain the real-time movement speed of the moving target.

[0023] In the process of environmental objects moving, children may also move. When the movements of the two are superimposed, it is difficult to identify the real movement of the object. Usually, the identification of stationary objects can be used and used as markers. Then, the speed of all objects can be obtained based on the markers as a reference. However, this requires the identification of stationary objects. Since children will continue to move, the markers will be constantly replaced as the children move. The identification of stationary objects requires a more complex algorithm to be set. This is because the movement of children is irregular and the changes of stationary objects in the image are also irregular. Therefore, the identification algorithm must be relatively complex, and when there is a problem with the identification, it is easy to cause deviations in all subsequent speed calculations. In this solution, considering that the movement of objects in the image is the superposition of children's movement and the actual movement of the object, its minimum value must be the superposition of children's movement and 0. Therefore, based on the minimum value of the obtained speed, the speed of the child can be obtained, and then the actual speed of the object can be obtained, and subsequent judgments can be made based on this.

[0024] Reference Figure 6 As shown, by analyzing the texture of each feature, based on the movement of objects in the environment, the behavior states of environmental objects and children are extracted, including: Based on the texture analysis of each feature, locate the facial area of ​​the moving target; When the moving target does not have a facial region, the speed reduction when the distance between the moving target and the child is less than the safe distance is taken as the behavior state of the environmental object; When the moving target has a facial region, locating key feature points in the facial region of the moving target, wherein the key feature points include eyes, mouth and eyebrows; Analyze the texture of the facial area and extract facial expression features; Compare the extracted facial expression features with the known expression database to identify the current facial expression of the moving target; Count the number of blinks per unit time, and determine the opening and closing state of the eyes based on the occlusion of the pupils, and summarize and output the eye state of the moving target; Extract and classify features of images of the hand area to identify hand movements of moving targets; Track hand movements in continuous image frames, analyze the continuity and changes of hand movements, and output the gestures of moving targets; The current facial expression, eye state and gesture action of the moving target are regarded as the behavior state of the environmental object; The same analysis steps as the motion target are used to obtain the child’s current facial expression, eye state, and gesture movement, which are used as the child’s behavioral state.

[0025] Perform texture analysis on the located facial area, extract facial expression features by calculating facial muscle deformation, texture change and other features. These features can reflect the emotional state of the moving target, such as joy, anger, sorrow, and happiness, and provide reliable data support for subsequent dangerous behavior warnings. Some objects are not living things but can move, so they also need to be identified. The main thing is to obtain their speed when they are close to children, and make judgments and settings in subsequent models. If the speed is still too high when they are close to children, there is a certain risk and an early warning is needed.

[0026] Reference Figure 7 As shown, the model constructed by the random forest algorithm is trained using a historical environmental monitoring data set containing various environmental scenarios to obtain a classification and recognition model that specifically includes: S101: Setting the historical environment monitoring database as a model training data set; S102: Randomly select a data subset that has not been selected from the historical environment monitoring database as the root node of the decision tree; S103: establishing a first leaf node at the root node according to different behavior states of the moving target as a splitting criterion; S104: establishing a second leaf node at the first leaf node based on the difference in the behavior status of the children as a splitting criterion; S105: Summarize the root node and all its leaf nodes and output them as a decision tree; S106: Determine whether corresponding decision trees are established for all data subsets in the historical environment monitoring database. If so, output all decision trees as classification and recognition models. If not, return to step S102.

[0027] The core of the random forest algorithm is to create multiple decision trees and let these trees predict samples independently. The final prediction result is based on the vote (for classification problems) or average (for regression problems) of the prediction results of all trees. When building each tree, the algorithm randomly extracts samples and features from the original data set. This randomness helps to improve the generalization ability of the model.

[0028] Furthermore, the present solution also proposes a computer-readable storage medium on which a computer-readable program is stored. When the computer-readable program is called, the above-mentioned children's smart watch danger warning method based on environmental recognition is executed.

[0029] It is understandable that the storage medium may be a magnetic medium, such as a floppy disk, a hard disk, or a magnetic tape; an optical medium, such as a DVD; or a semiconductor medium, such as a solid state drive (SSD).

[0030] In summary, the advantages of the present invention are: by setting the movement of objects in the detection environment, extracting the behavioral status of environmental objects and the behavioral status of children and obtaining a classification and recognition model, real-time monitoring and early warning of the status of children and their guardians can be achieved, driving safety can be improved, and the occurrence of traffic accidents can be reduced; by pre-processing images (such as denoising, enhancement and grayscale processing), and using advanced classification and recognition algorithms (such as random forest algorithms), accurate identification of children's safety status can be achieved under complex lighting and occlusion conditions.

[0031] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions only describe the principles of the present invention. The present invention may be subject to various changes and improvements without departing from the spirit and scope of the present invention. These changes and improvements fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the attached claims and their equivalents.

Claims

1. A children's smart watch danger warning method based on environment recognition, characterized in that: include: Install a miniature camera on the smartwatch and debug the camera, adjusting its angle and focal length parameters to ensure that the camera captures a full view of the child's surroundings; The camera captures continuous image sequences in the environment in real time, and performs denoising, enhancement and grayscale processing on the captured images in turn; Extract edge, texture and shape features from the preprocessed image; Based on the image frame difference method, the motion of objects in the environment is detected by performing image dynamic detection on the continuous image sequence in the environment captured by the camera in real time; Through texture analysis of each feature, based on the movement of objects in the environment, the behavior state of environmental objects and the behavior state of children are extracted; Using the random forest algorithm, the behavioral states of environmental objects are linked to dangerous situations of children; The model constructed by the random forest algorithm is trained using a historical environmental monitoring data set containing various environmental scenarios to obtain a classification and recognition model; The continuous image sequence in the environment captured by the camera in real time is input into the classification and recognition model, and the corresponding warning signal and prompt signal are output to the child and the child's guardian respectively.

2. According to the method of claim 1, the method is characterized in that: The camera captures a continuous image sequence in the environment in real time, and performs denoising, enhancement and grayscale processing on the captured images in sequence, specifically including: Use image processing library to capture live images from camera; Use a median filter to remove noise from the image and replace the value of each pixel in the continuous image sequence of the environment with the median of its neighborhood pixel value; By analyzing the pixel value distribution of the image, the pixel value of each pixel is redistributed to even out the contrast of the image; Multiply all pixel values ​​of the image by a preset constant to enhance the brightness of the image; Obtain the weighted coefficients of human eye sensitivity to red, green and blue; Substitute the red, green and blue color components of each pixel in the continuous image sequence in the environment into the weighted average formula to obtain the grayscale value of each pixel in the continuous image sequence in the environment; The weighted average formula is: , In the formula, is the gray value of the pixel at the i-th row and j-th column, are the weighted coefficients of human eye sensitivity to red, green and blue, respectively. They are the red, green and blue color components of the pixel in the i-th row and j-th column respectively.

3. A method for warning danger of children's smart watches based on environment recognition according to claim 2, characterized in that: The extracting edge, texture and shape features from the preprocessed image specifically includes: Get the third-order Laplace operator; Perform convolution operation on the third-order Laplace operator and the preprocessed image to obtain the new pixel value of each pixel; In the image after Laplace operation, the zero crossing points where the pixel value changes from positive to negative or from negative to positive are identified, and the zero crossing points are output as edge points of the image; Set the neighborhood of each pixel to 8 pixels within a 3×3 range centered on the pixel; Determine whether the grayscale value of each neighboring pixel is greater than or equal to the grayscale value of the central pixel. If so, output the binary bit of the neighboring pixel as 1; if not, output the binary bit of the neighboring pixel as 0; Combining 8 binary bits clockwise from the upper left corner, we get an 8-bit binary number. Output the decimal number corresponding to the binary number as the texture feature value of the pixel position; Calculate the frequency of occurrence of texture eigenvalues ​​in each image and construct a texture eigenvalue frequency histogram; Based on the figure surrounded by edge points, its area, perimeter and direction are output as shape features; The Laplace operator is: , In the formula, is the third-order Laplace operator; The convolution operation is: , In the formula, is the new pixel value of the pixel at the i-th row and j-th column.

4. A method for warning danger of children's smart watches based on environment recognition according to claim 3, characterized in that: The image frame difference method is based on performing image dynamic detection on a continuous image sequence in the environment captured by a camera in real time to detect the motion of objects in the environment, specifically including: Continuously capture two adjacent frames of images from a continuous image sequence of the environment captured by the camera in real time; For each pixel position of the two frames of images, calculate the difference in grayscale values; The minimum amplitude of pixel value change that can be recognized by human eyes is used as the preset difference threshold; Determine whether the difference in the grayscale value of each pixel is greater than a preset difference threshold. If so, output that the pixel is in motion; if not, output that the pixel belongs to a static background; Cluster all pixels with motion to form connected regions, and output each connected region as a moving target in the environment; Repeatedly capture subsequent adjacent images to perform image dynamic detection, track the motion trajectory of each moving target and calculate the real-time motion speed of each moving target.

5. The method for warning danger of a children's smart watch based on environment recognition according to claim 4 is characterized in that: The calculation of the real-time motion speed of each moving target comprises the following steps: According to the moving trajectory of the moving target, the initial moving speed of the moving target is calculated; The minimum value of the initial movement speed is used as the child's movement speed; The child's movement speed is reversely superimposed on the initial movement speed to obtain the real-time movement speed of the moving target.

6. A method for warning danger of children's smart watches based on environment recognition according to claim 5, characterized in that: The extraction of the behavior state of the environmental objects and the behavior state of the children based on the movement of the objects in the environment through texture analysis of each feature specifically includes: Based on the texture analysis of each feature, locate the facial area of ​​the moving target; When the moving target does not have a facial region, the speed reduction when the distance between the moving target and the child is less than the safe distance is taken as the behavior state of the environmental object; When the moving target has a facial region, locating key feature points in the facial region of the moving target, wherein the key feature points include eyes, mouth and eyebrows; Analyze the texture of the facial area and extract facial expression features; Compare the extracted facial expression features with the known expression database to identify the current facial expression of the moving target; Count the number of blinks per unit time, and determine the opening and closing state of the eyes based on the occlusion of the pupils, and summarize and output the eye state of the moving target; Extract and classify features of images of the hand area to identify hand movements of moving targets; Track hand movements in continuous image frames, analyze the continuity and changes of hand movements, and output the gestures of moving targets; The current facial expression, eye state and gesture action of the moving target are regarded as the behavior state of the environmental object; The same analysis steps as the motion target are used to obtain the child’s current facial expression, eye state, and gesture movement, which are used as the child’s behavioral state.

7. A method for warning danger of children's smart watches based on environment recognition according to claim 6, characterized in that: The model constructed by the random forest algorithm is trained using a historical environmental monitoring data set containing various environmental scenes to obtain a classification recognition model, which specifically includes: S101: Setting the historical environment monitoring database as a model training data set; S102: Randomly select a data subset that has not been selected from the historical environment monitoring database as the root node of the decision tree; S103: establishing a first leaf node at the root node according to different behavior states of the moving target as a splitting criterion; S104: establishing a second leaf node at the first leaf node based on the difference in the behavior status of the children as a splitting criterion; S105: Summarize the root node and all its leaf nodes and output them as a decision tree; S106: Determine whether corresponding decision trees are established for all data subsets in the historical environment monitoring database. If so, output all decision trees as classification and recognition models. If not, return to step S102.

Images