A method and system for visually identifying dangers for infant safety monitoring and protection

Abstract

The present invention discloses a hazard visual recognition method and system for infant safety monitoring and protection, which relates to the field of infant safety monitoring and protection, including: retrieving video and image data of infant activity places based on big data, and performing multi-dimensional labeling processing on elements in the data; pre-processing the labeled data and performing feature extraction, and training a visual recognition model combination based on the division of labor of visual recognition tasks; performing subject recognition on real-time monitoring images, and marking the position coordinates of all infant individuals in the monitoring images; analyzing the biological type recognition results and behavior identification results of the risk assessment area, and evaluating the biological safety threat level of the individual infant; identifying the dangerous area of ​​the site structure, evaluating the structural danger of the site where the individual infant is located, comprehensively evaluating the biological safety threat level of the individual infant and the structural danger of the site where the individual infant is located, and performing risk warning for infant safety monitoring.

Description

Technical Field

[0001] The present invention relates to the field of infant safety monitoring and protection, and in particular to a hazard visual recognition method and system for infant safety monitoring and protection. Background Art

[0002] The visual recognition system for baby safety monitoring is used to monitor the baby's status and environment in real time to ensure the baby's safety and health. Such systems usually integrate advanced computer vision technology and can automatically detect and identify various potential risks or abnormal situations. At present, the main use scenario is family baby safety monitoring and protection, which generally realizes baby safety monitoring through baby posture recognition and room foreign object recognition.

[0003] This approach has location limitations. When infants are in open places, the sources of threats they may face will also change. For example, in open activity venues such as nurseries and parks, infants are less likely to come into contact with dangerous items such as sharp blades and sockets. However, the safety threats to infants from other people, animals or the terrain of the venue will increase. If timely risk warnings are not issued for safety hazards to infants in open places, the health and safety of infants may be harmed.

[0004] Therefore, how to realize a system that can provide real-time safety monitoring and protection for infants based on visual recognition of dangers in open places and provide timely warnings of safety and health hazards that infants may encounter is an urgent problem that needs to be solved. Summary of the invention

[0005] In order to solve the above technical problems, a hazard visual identification method for infant safety monitoring and protection is provided. This technical solution solves the problems raised in the above background technology.

[0006] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0007] A hazard visual identification method for infant safety monitoring and protection, comprising:

[0008] Retrieve video and image data of infant activity sites based on big data, and perform multi-dimensional labeling on the elements in the data;

[0009] Extract features from the preprocessed marking data and train the visual recognition model combination based on the division of visual recognition tasks;

[0010] Perform subject recognition on real-time monitoring images and mark the location coordinates of all infants in the monitoring images;

[0011] Extract the background image features of the risk assessment area, analyze the biological type recognition results and behavior recognition results of the risk assessment area, and assess the biological safety threat level of the individual infant;

[0012] Extract the background image features of the entire monitoring area, identify the dangerous areas of the site structure, evaluate the structural danger of the site where the individual infant is located, comprehensively evaluate the individual infant's biosafety threat level and the structural danger of the site, and issue risk warnings for infant safety monitoring.

[0013] Preferably, the retrieving video and image data of baby activity places based on big data and performing multi-dimensional labeling processing on elements in the data specifically includes:

[0014] Retrieve video and image data of infant activity sites based on big data, and select high-definition data as original data;

[0015] Set up visual recognition division of labor categories, including biological type recognition, behavior recognition and site structure recognition;

[0016] Based on the visual recognition division of labor categories, the biological type dimension, behavioral identification dimension, and site structure dimension of the individual elements in the original data are manually labeled;

[0017] The dangerousness level of the labels with different dimensions is searched, and the dangerousness level is divided based on the search results.

[0018] Preferably, the feature extraction of the preprocessed marking data and the training of the visual recognition model combination based on the division of visual recognition tasks specifically include:

[0019] The data is processed by standardizing the size and pixels, and image enhancement and denoising are performed after the processing.

[0020] Select a pre-trained model based on a convolutional neural network to extract features from the pre-processed data, and mark the extracted feature vector as a shared feature vector;

[0021] Obtain visual division of labor categories, and train dedicated models for biological type recognition, behavior identification, and place structure recognition based on the image data corresponding to the division of labor categories and shared feature vectors and the corresponding labeling association relationships;

[0022] Select a set number of data as the test data set, and evaluate the performance of the trained special model based on the test data. If the recognition accuracy of the special model reaches the preset standard value, the special model will be combined and deployed to the system application.

[0023] If the recognition accuracy of the dedicated model does not reach the preset standard value, the dedicated model is retrained after data optimization and parameter tuning until the recognition accuracy reaches the preset standard value.

[0024] Preferably, the subject recognition of the real-time monitoring image and marking the position coordinates of all infant individuals in the monitoring image specifically includes:

[0025] Obtain real-time monitoring images of the baby safety monitoring protection site, pre-process the real-time monitoring images, and use the deployed combined model to perform visual recognition on the real-time monitoring images;

[0026] Obtain the recognition result of the biological type recognition model, and confirm the subject of the individual whose element recognition result in the image is a baby;

[0027] Based on the top-down projection of the real-time monitoring image, a plane rectangular coordinate system is established with the intersection point of the lower left edge of the image as the coordinate origin, and the specific coordinates of all infant individuals in the coordinate system are determined and marked as the monitoring subject coordinates;

[0028] Set the scope of the subject risk assessment area and accurately identify the risk assessment areas of all monitored subjects.

[0029] Preferably, extracting the background image features of the risk assessment area, analyzing the biological type recognition results and behavior identification results of the risk assessment area, and assessing the biological safety threat level of the individual infant specifically includes:

[0030] Obtain the risk assessment areas corresponding to all monitored subjects, extract the background image features of the subject risk assessment area, and perform accurate visual recognition of the background of each subject risk assessment area;

[0031] Conduct a separate analysis of the visual recognition results;

[0032] Obtain the background biological type recognition results of the risk assessment area, count the danger level, number of organisms, volume of organisms and distance information relative to the coordinate position of the monitored subject corresponding to the recognized biological type, and mark the statistical results as biological type security threat variables;

[0033] Obtain the background behavior identification results of the risk assessment area, count the corresponding danger level of the identified biological behavior and the distance relative to the subject's coordinate position, mark the statistical results as biological behavior security threat variables, and record the change information of the biological coordinate movement distance for the set number of identifications;

[0034] A comprehensive analysis of the bio-type safety threat variables and the bio-behavior safety threat variables was performed to calculate the bio-threat risk index, which was used to assess the level of bio-safety threat to individual infants. The specific expression is: ;

[0035] In the formula, is the biological threat risk index, , , are the biological type danger level, biological behavior danger level and biological volume of the jth organism identified, respectively. is the Euclidean distance between the kth recorded coordinates of the jth organism and the main body coordinates, l is the number of records of the set coordinate movement distance, and m is the number of identified organisms in the risk assessment area.

[0036] Preferably, the extracting of background image features of the entire monitoring area, identifying dangerous areas of the site structure, assessing the dangerousness of the site structure where the individual infant is located, comprehensively assessing the biological safety threat level of the individual infant and the dangerousness of the site structure where the individual infant is located, and providing risk warning for infant safety monitoring specifically includes:

[0037] Obtain the background image of the entire monitoring area, extract the background image features, accurately identify the background image of the monitoring area, and obtain the site structure recognition results within the monitoring area;

[0038] Based on the results of site structure identification, the dangerous area of ​​the site structure is screened. According to the distance between each infant and the dangerous area of ​​the site structure and the corresponding danger level of the area, the site structure hazard assessment coefficient is calculated to assess the site structure hazard of each infant. The formula for calculating the site structure hazard assessment coefficient is: ;

[0039] In the formula, is the site structure hazard assessment coefficient, is the danger level corresponding to the area, is the Euclidean distance of the jth infant individual relative to the i-th danger zone;

[0040] Comprehensively evaluate the biological threat level of individual infants and the structural hazard of the site, obtain the biological threat risk index and the site structure hazard assessment coefficient, weight the biological threat risk index and the site structure hazard assessment coefficient, and calculate the safety risk warning coefficient of each individual infant;

[0041] A safety risk warning coefficient threshold for individual infants is set, and infants whose safety risk warning coefficient is greater than or equal to the safety risk warning coefficient threshold are marked as monitored safety risk infants. The monitored safety risk infants are marked and safety risk warning signals are issued.

[0042] Furthermore, a hazard visual recognition system for infant safety monitoring and protection is proposed, which is used to implement the hazard visual recognition method for infant safety monitoring and protection as described above, including:

[0043] A data collection module, which retrieves video and image data of infant activity sites based on big data, performs multi-dimensional labeling of elements in the data in terms of biological type, biological behavior, and site structure, and retrieves the degree of danger of the marked labels, and classifies the danger levels based on the retrieval results;

[0044] A visual recognition model training module, which performs feature extraction after preprocessing the marked data, trains special models for biological type recognition, behavior recognition and site structure recognition based on the division of visual recognition tasks, and combines qualified special models and deploys them to system applications;

[0045] A biosafety threat assessment module, which extracts background image features of the subject risk assessment area, performs precise visual recognition and separate analysis of the background of each subject risk assessment area, calculates a biosafety threat risk index, and assesses the biosafety threat level of the individual infant;

[0046] A site structure hazard assessment module, wherein the site structure hazard assessment module calculates a site structure hazard assessment coefficient based on the site structure identification results obtained within the monitoring area and analyzes the individual positions of infants to assess the site structure hazard;

[0047] A monitoring safety risk warning module is provided, wherein the monitoring safety risk warning module calculates the safety risk warning coefficient of each individual infant, and marks the infants whose safety risk warning coefficient is greater than or equal to the safety risk warning coefficient threshold as infants with monitoring safety risks, marks the infants with monitoring safety risks and issues safety risk warning signals.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] Based on big data, video and image data of infant activity sites are retrieved, and elements in the data are labeled in multiple dimensions. After preprocessing the labeled data, feature extraction is performed, and a combination of visual recognition models is trained based on the division of visual recognition tasks. Multiple models handle their corresponding recognition tasks respectively, which helps to improve the efficiency and accuracy of recognition. Subject recognition is performed on real-time monitoring images, and the position coordinates of all infants in the monitoring images are marked. The results of biological type recognition and behavioral identification in the risk assessment area are analyzed to assess the level of biological safety threats to individual infants, which helps to understand the biological complexity and danger of the environment in which infants live. Identifying dangerous areas of the site structure and assessing the structural danger of the site where the infant is located helps to grasp the structural safety hazards of the site where the infant is located, comprehensively assess the level of biological safety threats to individual infants and the structural danger of the site where they are located, and conduct risk warnings for infant safety monitoring, realizing real-time safety monitoring and protection of infants based on dangerous visual recognition in open places. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flow chart of a hazard visual identification method for infant safety monitoring and protection according to the present invention;

[0051] Figure 2 This is a flowchart of the present invention for retrieving video and image data of baby activity places based on big data and performing multi-dimensional labeling processing on elements in the data;

[0052] Figure 3 The present invention is a flowchart of extracting features from pre-processed marking data and training a visual recognition model based on the division of visual recognition tasks;

[0053] Figure 4 The present invention performs subject recognition on real-time monitoring images and marks the position coordinates of all infant individuals in the monitoring images; flow chart;

[0054] Figure 5 A flow chart of extracting background image features of a risk assessment area, analyzing biological type recognition results and behavior identification results of the risk assessment area, and evaluating the biological safety threat level of an individual infant in the present invention;

[0055] Figure 6 A flow chart of the invention for comprehensively assessing the biological safety threat level of individual infants and the structural danger of the place where they are located, and providing risk warning for infant safety monitoring;

[0056] Figure 7 The present invention is a flow chart of a hazard visual identification system for infant safety monitoring and protection. DETAILED DESCRIPTION

[0057] 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.

[0058] Reference Figure 1 As shown, a hazard visual identification method for infant safety monitoring and protection,

[0059] Retrieve video and image data of infant activity sites based on big data, and perform multi-dimensional labeling on the elements in the data;

[0060] Extract features from the preprocessed marking data and train the visual recognition model combination based on the division of visual recognition tasks;

[0061] Perform subject recognition on real-time monitoring images and mark the location coordinates of all infants in the monitoring images;

[0062] Extract the background image features of the risk assessment area, analyze the biological type recognition results and behavior recognition results of the risk assessment area, and assess the biological safety threat level of the individual infant;

[0063] Extract the background image features of the entire monitoring area, identify the dangerous areas of the site structure, evaluate the structural danger of the site where the individual infant is located, comprehensively evaluate the individual infant's biosafety threat level and the structural danger of the site, and issue risk warnings for infant safety monitoring.

[0064] Reference Figure 2 As shown in the figure, based on big data, the video and image data of infant activity places are retrieved, and the elements in the data are multi-dimensionally labeled, which specifically includes:

[0065] Retrieve video and image data of infant activity sites based on big data, and select high-definition data as original data;

[0066] Set up visual recognition division of labor categories, including biological type recognition, behavior recognition and site structure recognition;

[0067] Based on the visual recognition division of labor categories, the biological type dimension, behavioral identification dimension, and site structure dimension of the individual elements in the original data are manually labeled;

[0068] The dangerousness level of the labels with different dimensions is searched, and the dangerousness level is divided based on the search results.

[0069] Among them, the danger level classification is to quantify the search results based on historical cases, relevant scientific recognition and the characteristics of this dimension. For example, the biological danger level can be divided into harmless level, low threat level, medium threat level and high threat level based on the comprehensive assessment of the species type, body shape, aggressiveness and injury cases of the organism. Similarly, the biological behavior danger level is also divided according to this standard. For example, sleeping, resting and other behaviors are classified as harmless, walking is classified as low threat, running and exercise are classified as medium danger, and smoking and fighting are classified as high danger. Each threat level is set with a corresponding assessment coefficient, which serves as a reference and assessment basis for subsequent calculations of infant safety risks.

[0070] Reference Figure 3 As shown in the figure, the pre-processed marking data is subjected to feature extraction, and the visual recognition model combination is trained based on the division of visual recognition tasks, specifically including:

[0071] The data is processed by standardizing the size and pixels, and image enhancement and denoising are performed after the processing.

[0072] Select a pre-trained model based on a convolutional neural network to extract features from the pre-processed data, and mark the extracted feature vector as a shared feature vector;

[0073] Obtain visual division of labor categories, and train dedicated models for biological type recognition, behavior identification, and place structure recognition based on the image data corresponding to the division of labor categories and shared feature vectors and the corresponding labeling association relationships;

[0074] Select a set number of data as the test data set, and evaluate the performance of the trained special model based on the test data. If the recognition accuracy of the special model reaches the preset standard value, the special model will be combined and deployed to the system application.

[0075] If the recognition accuracy of the dedicated model does not reach the preset standard value, the dedicated model is retrained after data optimization and parameter tuning until the recognition accuracy reaches the preset standard value.

[0076] Reference Figure 4 As shown, subject recognition is performed on the real-time monitoring image, and the position coordinates of all infant individuals in the monitoring image are marked, specifically including:

[0077] Obtain real-time monitoring images of the baby safety monitoring protection site, pre-process the real-time monitoring images, and use the deployed combined model to perform visual recognition on the real-time monitoring images;

[0078] Obtain the recognition result of the biological type recognition model, and confirm the subject of the individual whose element recognition result in the image is a baby;

[0079] Based on the top-down projection of the real-time monitoring image, a plane rectangular coordinate system is established with the intersection point of the lower left edge of the image as the coordinate origin, and the specific coordinates of all infant individuals in the coordinate system are determined and marked as the monitoring subject coordinates;

[0080] Set the scope of the subject risk assessment area and accurately identify the risk assessment areas of all monitored subjects.

[0081] Based on subject recognition, the positions of all infants in the monitoring area are first calibrated, and then a separate risk analysis and assessment is performed on each individual. The scope of the subject risk assessment area needs to be comprehensively set after considering the actual size of the monitoring range, the number of infants, and the element complexity of the monitoring image.

[0082] Reference Figure 5 As shown, the background image features of the risk assessment area are extracted, and the biological type recognition results and behavior recognition results of the risk assessment area are analyzed to assess the biological safety threat level of the individual infant, including:

[0083] Obtain the risk assessment areas corresponding to all monitored subjects, extract the background image features of the subject risk assessment area, and perform accurate visual recognition of the background of each subject risk assessment area;

[0084] Conduct a separate analysis of the visual recognition results;

[0085] Obtain the background biological type recognition results of the risk assessment area, count the danger level, number of organisms, volume of organisms and distance information relative to the coordinate position of the monitored subject corresponding to the recognized biological type, and mark the statistical results as biological type security threat variables;

[0086] Obtain the background behavior identification results of the risk assessment area, count the corresponding danger level of the identified biological behavior and the distance relative to the subject's coordinate position, mark the statistical results as biological behavior security threat variables, and record the change information of the biological coordinate movement distance for the set number of identifications;

[0087] A comprehensive analysis of the bio-type safety threat variables and the bio-behavior safety threat variables was performed to calculate the bio-threat risk index, which was used to assess the level of bio-safety threat to individual infants. The specific expression is: ;

[0088] In the formula, is the biological threat risk index, , , are the biological type danger level, biological behavior danger level and biological volume of the jth organism identified, respectively. is the Euclidean distance between the kth recorded coordinates of the jth organism and the main body coordinates, l is the number of records of the set coordinate movement distance, and m is the number of identified organisms in the risk assessment area.

[0089] By combining the hazard of biological species with the hazard of biological behavior for comprehensive assessment, the focus on multiple factors of biological hazard is emphasized, which increases the credibility of the assessment results. The calculation method is , where x and y represent the marked coordinates of the current infant subject respectively. The movement speed and movement trend of the threatening organism can be evaluated through the change information of the biological coordinate movement distance. When the coordinate movement distance is large each time, it means that the biological movement speed is fast and the safety threat to the infant is higher.

[0090] Reference Figure 6 As shown, the background image features of the entire monitoring area are extracted, the dangerous areas of the site structure are identified, the dangerousness of the site structure where the individual infant is located is evaluated, the level of biological safety threat to the individual infant and the dangerousness of the site structure are comprehensively evaluated, and risk warnings for infant safety monitoring are carried out. Specifically, the following are included:

[0091] Obtain the background image of the entire monitoring area, extract the background image features, accurately identify the background image of the monitoring area, and obtain the site structure recognition results within the monitoring area;

[0092] Based on the results of site structure identification, the dangerous area of ​​the site structure is screened. According to the distance between each infant and the dangerous area of ​​the site structure and the corresponding danger level of the area, the site structure hazard assessment coefficient is calculated to assess the site structure hazard of each infant. The formula for calculating the site structure hazard assessment coefficient is: ;

[0093] In the formula, is the site structure hazard assessment coefficient, is the danger level corresponding to the area, is the Euclidean distance of the jth infant individual relative to the i-th danger zone;

[0094] The dangerous areas of the site structure include stairs, slopes and other structures that have a greater impact on the safety of infants. The classification of the dangerous areas is based on the results of searching historical accident cases and factors such as actual height differences. The dangerous areas of the site structure are related to the nature of the site. For example, the terrain slopes and high and low drop areas in the park are significantly more than those in the nursery or family venues, and the site structure is also more dangerous.

[0095] Comprehensively evaluate the biological threat level of individual infants and the structural hazard of the site, obtain the biological threat risk index and the site structure hazard assessment coefficient, weight the biological threat risk index and the site structure hazard assessment coefficient, and calculate the safety risk warning coefficient of each individual infant. , the calculation expression of safety risk warning coefficient is: ;

[0096] In the formula , They are the weight coefficients of biological threat risk index and site structure hazard assessment coefficient respectively.

[0097] A safety risk warning coefficient threshold for individual infants is set, and infants whose safety risk warning coefficient is greater than or equal to the safety risk warning coefficient threshold are marked as monitored safety risk infants. The monitored safety risk infants are marked and safety risk warning signals are issued.

[0098] The threshold value of the safety risk warning coefficient for individual infants is set by professionals in this field after comprehensively considering the safety risk warning coefficient, visual recognition accuracy, the nature of the monitoring location, and the infant monitoring protection safety level, and will not be repeated here.

[0099] For further information, see Figure 7 As shown, based on the same inventive concept as the above-mentioned hazard visual identification method for infant safety monitoring and protection, the present solution also proposes a hazard visual identification system for infant safety monitoring and protection, comprising:

[0100] A data collection module, which retrieves video and image data of infant activity sites based on big data, performs multi-dimensional labeling of elements in the data in terms of biological type, biological behavior, and site structure, and retrieves the degree of danger of the marked labels, and classifies the danger levels based on the retrieval results;

[0101] A visual recognition model training module, which performs feature extraction after preprocessing the marked data, trains special models for biological type recognition, behavior recognition and site structure recognition based on the division of visual recognition tasks, and combines qualified special models and deploys them to system applications;

[0102] A biosafety threat assessment module, which extracts background image features of the subject risk assessment area, performs precise visual recognition and separate analysis of the background of each subject risk assessment area, calculates a biosafety threat risk index, and assesses the biosafety threat level of the individual infant;

[0103] A site structure hazard assessment module, wherein the site structure hazard assessment module calculates a site structure hazard assessment coefficient based on the site structure identification results obtained within the monitoring area and analyzes the individual positions of infants to assess the site structure hazard;

[0104] A monitoring safety risk warning module is provided, wherein the monitoring safety risk warning module calculates the safety risk warning coefficient of each individual infant, and marks the infants whose safety risk warning coefficient is greater than or equal to the safety risk warning coefficient threshold as infants with monitoring safety risks, marks the infants with monitoring safety risks and issues safety risk warning signals.

[0105] Furthermore, the present solution also proposes a storage medium for a hazard visual identification method for infant safety monitoring and protection, on which a computer-readable program is stored. When the computer-readable program is called, the above-mentioned hazard visual identification method for infant safety monitoring and protection is executed.

[0106] 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).

[0107] 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 method for visually identifying dangers for infant safety monitoring and protection, characterized in that: include: Retrieve video and image data of infant activity sites based on big data, and perform multi-dimensional labeling on the elements in the data; Extract features from the preprocessed marking data and train the visual recognition model combination based on the division of visual recognition tasks; Perform subject recognition on real-time monitoring images and mark the location coordinates of all infants in the monitoring images; Extract the background image features of the risk assessment area, analyze the biological type recognition results and behavior recognition results of the risk assessment area, and assess the biological safety threat level of the individual infant; Extract the background image features of the entire monitoring area, identify the dangerous areas of the site structure, evaluate the dangerousness of the site structure where the individual infant is located, comprehensively evaluate the biological safety threat level of the individual infant and the dangerousness of the site structure, and provide risk warning for infant safety monitoring; The extracting of background image features of the risk assessment area, analyzing the biological type recognition results and behavior identification results of the risk assessment area, and assessing the biological safety threat level of the individual infant specifically include: Obtain the risk assessment areas corresponding to all monitored subjects, extract the background image features of the subject risk assessment area, and perform accurate visual recognition of the background of each subject risk assessment area; Conduct a separate analysis of the visual recognition results; Obtain the background biological type recognition results of the risk assessment area, count the danger level, number of organisms, volume of organisms and distance information relative to the coordinate position of the monitored subject corresponding to the recognized biological type, and mark the statistical results as biological type security threat variables; Obtain the background behavior identification results of the risk assessment area, count the corresponding danger level of the identified biological behavior and the distance relative to the subject's coordinate position, mark the statistical results as biological behavior security threat variables, and record the change information of the biological coordinate movement distance for the set number of identifications; A comprehensive analysis of the bio-type safety threat variables and the bio-behavior safety threat variables was performed to calculate the bio-threat risk index, which was used to assess the level of bio-safety threat to individual infants. The specific expression is: ; In the formula, is the biological threat risk index, , , are the biological type danger level, biological behavior danger level and biological volume of the jth organism identified, respectively. is the Euclidean distance between the kth recorded coordinates of the jth organism and the main body coordinates, l is the number of records of the set coordinate movement distance, and m is the number of identified organisms in the risk assessment area; The extracting of background image features of the entire monitoring area, identifying dangerous areas of the site structure, assessing the dangerousness of the site structure where the individual infant is located, comprehensively assessing the biological safety threat level of the individual infant and the dangerousness of the site structure where the individual infant is located, and providing risk warning for infant safety monitoring specifically include: Obtain the background image of the entire monitoring area, extract the background image features, accurately identify the background image of the monitoring area, and obtain the site structure recognition results within the monitoring area; Based on the results of site structure identification, the dangerous area of ​​the site structure is screened. According to the distance between each infant and the dangerous area of ​​the site structure and the corresponding danger level of the area, the site structure hazard assessment coefficient is calculated to assess the site structure hazard of each infant. The formula for calculating the site structure hazard assessment coefficient is: ; In the formula, is the site structure hazard assessment coefficient, is the danger level corresponding to the area, is the Euclidean distance of the jth infant individual relative to the i-th danger zone; Comprehensively evaluate the biological threat level of individual infants and the structural hazard of the site, obtain the biological threat risk index and the site structure hazard assessment coefficient, weight the biological threat risk index and the site structure hazard assessment coefficient, and calculate the safety risk warning coefficient of each individual infant; A safety risk warning coefficient threshold for individual infants is set, and infants whose safety risk warning coefficient is greater than or equal to the safety risk warning coefficient threshold are marked as monitored safety risk infants. The monitored safety risk infants are marked and safety risk warning signals are issued.

2. A method for visually identifying dangers for infant safety monitoring and protection according to claim 1, characterized in that: The method of retrieving the video and image data of the baby activity place based on big data and performing multi-dimensional labeling processing on the elements in the data specifically includes: Retrieve video and image data of infant activity sites based on big data, and select high-definition data as original data; Set up visual recognition division of labor categories, including biological type recognition, behavior recognition and site structure recognition; Based on the visual recognition division of labor categories, the biological type dimension, behavioral identification dimension, and site structure dimension of the individual elements in the original data are manually labeled; The dangerousness level of the labels with different dimensions is searched, and the dangerousness level is divided based on the search results.

3. A method for visually identifying dangers for infant safety monitoring and protection according to claim 2, characterized in that: The feature extraction of the pre-processed marking data and the training of the visual recognition model combination based on the division of visual recognition tasks specifically include: The data is processed by standardizing the size and pixels, and image enhancement and denoising are performed after the processing. Select a pre-trained model based on a convolutional neural network to extract features from the pre-processed data, and mark the extracted feature vector as a shared feature vector; Obtain visual division of labor categories, and train dedicated models for biological type recognition, behavior identification, and place structure recognition based on the image data corresponding to the division of labor categories and shared feature vectors and the corresponding labeling association relationships; Select a set number of data as the test data set, and evaluate the performance of the trained special model based on the test data. If the recognition accuracy of the special model reaches the preset standard value, the special model will be combined and deployed to the system application. If the recognition accuracy of the dedicated model does not reach the preset standard value, the dedicated model is retrained after data optimization and parameter tuning until the recognition accuracy reaches the preset standard value.

4. A method for visually identifying dangers for infant safety monitoring and protection according to claim 3, characterized in that: The subject recognition of the real-time monitoring image and marking the position coordinates of all infant individuals in the monitoring image specifically includes: Obtain real-time monitoring images of the baby safety monitoring protection site, pre-process the real-time monitoring images, and use the deployed combined model to perform visual recognition on the real-time monitoring images; Obtain the recognition result of the biological type recognition model, and confirm the subject of the individual whose element recognition result in the image is a baby; Based on the top-down projection of the real-time monitoring image, a plane rectangular coordinate system is established with the intersection point of the lower left edge of the image as the coordinate origin, and the specific coordinates of all infant individuals in the coordinate system are determined and marked as the monitoring subject coordinates; Set the scope of the subject risk assessment area and accurately identify the risk assessment areas of all monitored subjects.

5. A hazard visual identification system for infant safety monitoring and protection, used to implement a hazard visual identification method for infant safety monitoring and protection as claimed in any one of claims 1 to 4, characterized in that: include: A data collection module, which retrieves video and image data of infant activity sites based on big data, performs multi-dimensional labeling of elements in the data in terms of biological type, biological behavior, and site structure, and retrieves the degree of danger of the marked labels, and classifies the danger levels based on the retrieval results; A visual recognition model training module, which performs feature extraction after preprocessing the marked data, trains special models for biological type recognition, behavior recognition and site structure recognition based on the division of visual recognition tasks, and combines qualified special models and deploys them to system applications; A biosafety threat assessment module, which extracts background image features of the subject risk assessment area, performs precise visual recognition and separate analysis of the background of each subject risk assessment area, calculates a biosafety threat risk index, and assesses the biosafety threat level of the individual infant; A site structure hazard assessment module, wherein the site structure hazard assessment module calculates a site structure hazard assessment coefficient based on the site structure identification results obtained within the monitoring area and analyzes the individual positions of infants to assess the site structure hazard; A monitoring safety risk warning module is provided, wherein the monitoring safety risk warning module calculates the safety risk warning coefficient of each individual infant, and marks the infants whose safety risk warning coefficient is greater than or equal to the safety risk warning coefficient threshold as infants with monitoring safety risks, marks the infants with monitoring safety risks and issues safety risk warning signals.

Images