First aid resource information early warning method and system based on AI recognition

Abstract

The application belongs to the technical field of image processing, and particularly relates to an emergency resource information early warning method and system based on AI recognition, which comprises the following steps: establishing a monitoring scene coordinate system, calibrating operation key points of emergency facilities, and setting an area of interest; collecting video streams in real time, extracting foreground targets, and performing attribute preliminary screening according to morphological rules to determine a normalization adjustment coefficient; constructing a spatial obstruction potential energy model containing distance potential energy based on the distance from the center of mass of the foreground target to the operation key points, and calculating an instantaneous spatial obstruction index in combination with the normalization adjustment coefficient; utilizing dynamic and static mode decoupling logic, calculating a risk confidence according to the fluctuation change rate of the instantaneous spatial obstruction index, and triggering an early warning when the risk confidence exceeds a threshold value. By introducing a logarithmic potential field and a stability weight factor, the application realizes accurate assessment of the distance hazard of an obstruction and effective filtering of dynamic interference, and significantly improves the accuracy of emergency resource obstruction early warning.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing. More particularly, the present application relates to an emergency resource information early warning method and system based on AI recognition. BACKGROUND

[0002] With the construction of urban public safety system, AED automatic external defibrillator cabinet, miniature fire station, wall-mounted first aid kit and other emergency facilities are increasingly widely deployed in public places. In order to ensure that these facilities can be smoothly taken in emergency situations, prevent them from being blocked by goods, vehicles and other sundries, it is necessary to monitor the surrounding environment in real time.

[0003] The current mainstream emergency facility monitoring solution in the industry mainly relies on fixed area background difference method or general target detection algorithm such as YOLO series. Its core logic is that when foreign matter appears in the designated monitoring area (ROI), and the pixel change area ratio exceeds the set threshold and lasts for a certain time, the system determines that the channel is blocked and triggers an alarm.

[0004] However, the existing technology has obvious limitations in complex urban public environment. Due to the lack of consideration of spatial potential, only the blocking area is calculated, which cannot identify the specific impact of object position on the use of emergency facilities, for example, a small sticker close to the handle may be more harmful than a large box far from the cabinet, resulting in a high risk of false negatives; secondly, the existing technology is prone to confusion between dynamic and static modes. In places with dense pedestrian flow, natural shaking caused by pedestrians leaning, talking on the phone or cleaners wiping the equipment will cause continuous pixel changes, which is easy to be misjudged by the system as object stacking, causing a large number of invalid work orders; in addition, the existing technology lacks shape logic, making it difficult to distinguish between regular objects such as express boxes and irregular human bodies, resulting in a lack of pertinence in monitoring. SUMMARY

[0005] To solve the technical problems that the existing technology cannot effectively distinguish the harmfulness of object blocking position to use and is prone to confusion between human body dynamic stay and object static stacking, resulting in high false positive rate, the present application provides solutions in the following aspects.

[0006] In a first aspect, the present application provides an AI recognition-based first-aid resource information early warning method, comprising: establishing a monitoring scene coordinate system, marking the center point of the operating handle of the first-aid facility in the image as a potential field singularity, and dividing the physical contour area of the first-aid cabinet into a region of interest, while collecting the average brightness value of the environment as the environmental light reference; real-time access to the video stream, using the mixed Gaussian background modeling algorithm to remove the background to extract the foreground moving target, performing morphological processing on the foreground moving target to form a complete connected domain, and calculating the real volume ratio of the connected domain, and performing attribute preliminary screening on the foreground moving target according to the real volume ratio to determine the normalized adjustment coefficient; calculating the geometric centroid of the connected domain, according to the straight-line distance from the geometric centroid to the center point of the operating handle, the normalized adjustment coefficient and the pixel area of the connected domain, using a preset logarithmic potential field model to calculate the instantaneous spatial obstruction index generated by the object in the current frame, wherein the instantaneous spatial obstruction index is positively correlated with the normalized adjustment coefficient and the pixel area, and is negatively correlated with the straight-line distance; calculating the fluctuation change rate according to the instantaneous spatial obstruction index of the current frame and the last frame, using the fluctuation change rate to construct a stability weight factor, combining the risk confidence accumulated at the last moment and the natural decay constant to iteratively calculate the risk confidence at the current moment, and performing graded early warning when the risk confidence exceeds the preset threshold.

[0007] By constructing a logarithmic model containing distance potential, the present application can assign different obstruction weights according to the distance between the object and the key operation point, realize spatial potential evaluation of closer distance and greater harm, and effectively solve the defect that the traditional area threshold method cannot evaluate the position harm.

[0008] Preferably, the attribute preliminary screening of the foreground moving target according to the real volume ratio to determine the normalized adjustment coefficient comprises: calculating the real volume ratio of the connected domain, the real volume ratio being the ratio of the area of the connected domain to the area of its convex hull; judging whether the real volume ratio is lower than a preset real volume ratio threshold, if lower than the preset real volume ratio threshold, marking as a low-risk suspected target, and adjusting the value of the normalized adjustment coefficient downward; if not lower than the preset real volume ratio threshold, marking as a high-risk suspected target, and keeping or adjusting the value of the normalized adjustment coefficient upward.

[0009] Preferably, the use of a preset logarithmic potential field model to calculate the instantaneous spatial obstruction index generated by the object in the current frame comprises: calculating the ratio of the potential energy radiation radius and the distance factor, and taking the natural logarithm of the ratio after adding 1 to obtain the distance attenuation factor; wherein the distance factor is composed of the sum of the pixel distance from the object centroid to the handle of the first-aid cabinet and the safety buffer constant; multiplying the normalized adjustment coefficient, the pixel area of the foreground connected domain and the distance attenuation factor to obtain the instantaneous spatial obstruction index.

[0010] The natural logarithm function is used to represent the physical field strength attenuation characteristics, so that the hindering index smoothly rises when the object is close to the handle, and tends to zero when it is far away, and the calculation error of zero denominator is prevented by the safety buffer constant.

[0011] Preferably, the normalization adjustment coefficient is dynamically adjusted according to the total resolution of the image and the regularity of the shape of the foreground moving target, and the regularity is represented by the solid volume ratio, so as to suppress the weight of irregular objects.

[0012] By preliminary screening of the target through the regularity and dynamically adjusting the coefficient, prior knowledge can be introduced in the early stage, and the calculation weight of suspected irregular targets such as human bodies is automatically reduced, thereby further reducing the false alarm rate caused by short-term stay or operation of personnel from the source.

[0013] Preferably, the fluctuation change rate is calculated according to the instantaneous spatial hindering index of the current frame and the previous frame, comprising: reading the instantaneous spatial hindering index calculated in the previous frame and the instantaneous spatial hindering index of the current frame; calculating the absolute value of the difference between the instantaneous spatial hindering index of the current frame and the instantaneous spatial hindering index of the previous frame as the fluctuation change rate.

[0014] Preferably, the risk confidence of the current time is iteratively calculated, comprising: calculating a stability weight factor using an exponential function, wherein the exponential part of the stability weight factor is the negative value of the product of a stability sensitivity coefficient and the fluctuation change rate; multiplying the instantaneous spatial hindering index of the current frame by the stability weight factor to obtain a risk increment of the current frame; adding the risk increment of the current frame to the risk confidence accumulated at the previous time, and subtracting the natural attenuation constant to obtain the final risk confidence accumulated at the current time.

[0015] By introducing the stability weight factor in the form of an exponential function, the fluctuation change rate of dynamic targets such as pedestrians is large, the weight factor tends to zero, and the hindering index is not accumulated; while for static objects, the fluctuation change rate is very small, the weight factor tends to 1, and the risk confidence is quickly accumulated, thereby realizing accurate early warning.

[0016] Preferably, in the calculation of the risk confidence, the boundary condition limiting step is further included: if the calculated risk confidence is less than zero, the risk confidence is forcibly assigned to zero; if the calculated risk confidence is greater than a set saturation upper limit, the risk confidence is forcibly assigned to the saturation upper limit.

[0017] Preferably, the executing hierarchical early warning comprises: when the risk confidence is greater than a first preset proportion threshold and continuously increases, triggering a first-level early warning, playing a voice prompt; when the risk confidence is greater than a second preset proportion threshold, determining as a substantial obstruction, triggering a second-level alarm, intercepting a current frame picture and superimposing a heat map of the risk confidence and the instantaneous spatial obstruction index, and pushing to a management cloud platform.

[0018] The hierarchical early warning mechanism is adopted, slight or initial risks are only locally prompted by voice, unintended obstruction behaviors are effectively discouraged, and for serious risks, remote alarms are triggered and visual heat map evidences are pushed, so that management personnel can quickly research and judge and take evidence, and the efficiency and pertinence of emergency management are improved.

[0019] Preferably, the straight-line distance adopts the Euclidean distance, and the operation handle center point is taken as a singular point of the potential energy field, which is the highest protection level area.

[0020] In a second aspect, the present application provides an emergency resource information early warning system based on AI recognition, comprising a processor and a memory, the memory stores computer program instructions, when the computer program instructions are executed by the processor, the above-mentioned one kind of emergency resource information early warning method based on AI recognition is realized.

[0021] By adopting the above technical solution, the above-mentioned one kind of emergency resource information early warning method based on AI recognition is generated into a computer program and stored in the memory, so as to be loaded and executed by the processor, so as to manufacture a terminal device according to the memory and the processor, and facilitate use.

[0022] The present application maps the physical position of the obstructing object into a nonlinear spatial obstruction index by constructing a logarithmic potential energy field model based on the distance between the geometric centroid and the operation key point, automatically gives different weights according to the distance between the object and the operation part of the emergency facility, accurately evaluates the actual obstruction degree of the obstruction in different positions to the facility taking, solves the problem that the traditional area threshold method cannot evaluate the position hazard, and effectively realizes the key protection of the key operation area.

[0023] Further, the system automatically suppresses the risk accumulation of the human target with a small action feature by monitoring the fluctuation rate of the instantaneous obstruction index, and quickly responds to the obstruction behavior of the stationary object, effectively eliminates the false alarm caused by the short-term stay of personnel, and significantly improves the robustness of the early warning system while ensuring the sensitive detection of the illegal stacking behavior. BRIEF DESCRIPTION OF DRAWINGS

[0024] The above and other objects, features and advantages of the present application will become more apparent from the following detailed description read in conjunction with the accompanying drawings, in which several embodiments of the application are illustrated by way of example. In the drawings:

[0025] Figure 1 is a flow chart schematically showing an AI recognition-based first-aid resource information early warning method in the present application;

[0026] Figure 2 is a schematic diagram of a risk accumulation response mechanism in an embodiment of the present application;

[0027] Figure 3 is a distribution model diagram of an operation hindrance potential field of a first-aid resource. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the present application will be apparently and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the protection scope of the present application.

[0029] The specific embodiments of the present application will be described in detail below in conjunction with the drawings.

[0030] The embodiments of the present application disclose an AI recognition-based first-aid resource information early warning method, referring to Figure 1 , comprising steps S1-S4:

[0031] S1, establishing a monitoring scene coordinate system, marking the operation handle center point of the first-aid facility as a potential field singularity in an image, and dividing the physical outline region of the first-aid cabinet as a region of interest, while collecting the average brightness value of the environment as an ambient light reference.

[0032] In an optional embodiment, the base image of the first-aid facility can be collected by a front-end camera, and a two-dimensional pixel coordinate system is constructed in an image processing unit. Specifically, the operation handle center point of the first-aid facility is marked in the image, which is the singularity of the potential field, i.e., the region with the highest protection level. At the same time, the physical outline region of the first-aid cabinet is set as the ROI region of interest, and the average brightness value of the current environment is collected as the ambient light reference for subsequent removal of false triggers caused by sudden changes in light.

[0033] For example, in a 1920x1080 resolution monitoring screen, the first-aid cabinet is located in the center of the screen. After calibration, the coordinates of the operation handle center point are (960, 540), and the ROI region is set as a rectangular frame covering the entire front face of the first-aid cabinet.

[0034] Thus, by accurately calibrating the handle, the key point, the foundation is laid for the subsequent construction of the spatial potential field with this point as the core, so that the system can perceive the importance difference of different positions.

[0035] S2, real-time access to video stream, using mixed Gaussian background modeling algorithm to remove the background to extract the foreground moving target, morphological processing is carried out on the foreground moving target to form a complete connected domain, and the real volume ratio of the connected domain is calculated. According to the real volume ratio, the attribute preliminary screening of the foreground moving target is carried out to determine the normalization adjustment coefficient.

[0036] In an optional embodiment, real-time access to video stream, and processing each frame image, using mixed Gaussian background modeling (GMM) algorithm to remove the background, thereby extracting the foreground moving target, and performing dilation and closing operation on the foreground target to fill the internal holes and connect the broken edges of the target, forming a complete connected domain.

[0037] Further, morphological rule degree preliminary screening filtering is carried out, specifically, the real volume ratio of the connected domain is calculated, that is, the ratio of the connected domain area to its convex hull area. The logic is that due to the irregular shape, the human body usually has a lower real volume ratio, while the boxes, vehicles and other illegal stacking objects usually have a higher real volume ratio. By setting a preset real volume ratio threshold, for example, 0.5, the target with a real volume ratio lower than the threshold is marked as a low-risk suspected target, and its subsequent weight coefficient (normalization adjustment coefficient) is reduced; while the target with a real volume ratio higher than the threshold is marked as a high-risk suspected target, and the corresponding normalization adjustment coefficient is kept or increased.

[0038] For example, in a certain detection, a foreground connected domain is extracted, with a pixel area of 5000 and a calculated convex hull area of 8000, so the real volume ratio is 5000 / 8000=0.625. Since 0.625>0.5, the system marks it as a high-risk suspected target, and speculates that it may be a regular object such as a box, and sets a higher normalization adjustment coefficient; if another connected domain is extracted with a real volume ratio of 0.4, it is speculated to be a human body leaning against, and is marked as a low-risk suspected target, and the corresponding normalization adjustment coefficient is reduced.

[0039] Thus, by classifying the target through morphological features and determining the normalization adjustment coefficient accordingly, prior knowledge can be introduced at an early stage to reduce the possibility of false positives of non-rigid objects such as human bodies in subsequent calculations.

[0040] S3, calculate the geometric centroid of the connected domain, according to the straight line distance from the geometric centroid to the center point of the operating handle, a normalization adjustment coefficient, and the pixel area of the connected domain, calculate the instantaneous spatial obstruction index generated by the object in the current frame by using a preset logarithmic potential field model, wherein the instantaneous spatial obstruction index is positively correlated with the normalization adjustment coefficient and the pixel area, and is negatively correlated with the straight line distance.

[0041] In an optional embodiment, the total pixel number of the foreground connected domain can be acquired, and the geometric centroid of the connected domain is calculated. The straight line distance from the centroid to the center point of the operating handle is calculated by using the Euclidean distance formula, and the instantaneous spatial obstruction index generated by the object in the current frame is calculated by using a preset logarithmic potential field model in combination with a normalization adjustment coefficient and the pixel area of the connected domain, and the instantaneous spatial obstruction index satisfies the relationship:

[0042]

[0043] wherein, is the instantaneous spatial obstruction index generated by the object in the current frame calculated; is the pixel area of the foreground connected domain; is the pixel distance from the centroid of the object to the handle of the first-aid cabinet; is a safety buffer constant, and an exemplary value is 20. The safety buffer constant prevents the program from crashing when the object is close to the handle, and adjusts the sensitivity saturation curve of the potential field at an extremely close distance; is a potential energy radiation radius, and an exemplary value is 1 / 3 of the diagonal length of the image, which defines the effective influence range of the potential field; is a natural logarithm function. The logarithm function is used to represent the physical field strength attenuation characteristic. When decreases, the fractional term increases, and the logarithmic value smoothly rises. The addition of 1 is to ensure that when the object is at an infinite distance, the instantaneous spatial obstruction index is 0. is a normalization adjustment coefficient, which is dynamically adjusted according to the total resolution of the image and the morphological regularity.

[0044] Exemplarily, it is assumed that the object area , when the object is close to the handle, the centroid distance of the object to the handle is pixels, and ; when the object is away from the handle, the centroid distance of the object to the handle is pixels, . It can be seen that although the object areas are the same, due to the different distances, the instantaneous spatial obstruction index generated by the object close to the handle is much higher than that generated by the object away from the handle.

[0045] ​Thus, by constructing a logarithmic potential field model, the degree of obstruction of an object can be non-linearly evaluated according to the distance between the object and the key part, ensuring the key area is protected, which meets the actual needs of the use of first-aid facilities.

[0046] S4, calculate the fluctuation change rate according to the instantaneous spatial obstruction index of the current frame and the previous frame, construct a stability weight factor using the fluctuation change rate, combine the risk confidence accumulated at the previous time and the natural decay constant to iteratively calculate the risk confidence at the current time, and execute a graded early warning when the risk confidence exceeds a preset threshold.

[0047] In an optional embodiment, a buffer queue can be set, the instantaneous spatial obstruction index calculated in the previous frame and the instantaneous spatial obstruction index of the current frame are read, and the absolute value of the difference between the instantaneous spatial obstruction index of the current frame and the instantaneous spatial obstruction index of the previous frame is taken as the fluctuation change rate, which is used to reflect the stability of the occlusion.

[0048] Further, a stability weight factor can be constructed using the fluctuation change rate, the risk confidence accumulated at the previous time and the natural decay constant are combined to iteratively calculate the risk confidence at the current time, and the iterative accumulation relationship of the risk confidence is:

[0049]

[0050] wherein, is the final risk confidence accumulated at the current time, which is the only basis for the system to determine whether to alarm; is the risk confidence accumulated at the previous time; is the stability weight factor, which is between 0 and 1, is the stability sensitivity coefficient, which is exemplarily 5; is the calculated fluctuation change rate; is the natural decay constant, which is exemplarily 0.5, if the occlusion is removed, the system needs to automatically fall back to 0 through the accumulated risk value.

[0051] In the optional embodiment, in order to prevent numerical abnormalities in the formula from occurring in a long time, the following boundary condition limiting steps can be performed: if the calculated risk confidence is less than zero, the risk confidence is forcibly assigned to zero; if the calculated risk confidence is greater than a set saturation upper limit, exemplarily, such as a saturation upper limit of 100, the risk confidence is forcibly assigned to the saturation upper limit.

[0052] Exemplarily, let be 30, the previous time , if it is a static object such as a box, the object is static, extremely stable, and suppose ​The stability weight factor is , It can be seen that the risk confidence rises rapidly from 10 to 38, and the system identifies as high risk accumulation; if it is a dynamic interference such as a pedestrian, the person is in front of the cabinet, but the body has a swing, The fluctuation is large, and it is assumed that The stability weight factor is , It can be seen that although is large, due to the large fluctuation, the weight is extremely low, and the risk confidence decreases instead.

[0053] Further, the value of the real-time monitoring can be graded and warned, when the risk confidence is greater than a first preset proportion threshold such as 30 and continuously rises, triggering a first-level warning, playing a voice prompt "please do not block the emergency facility"; when the risk confidence is greater than a second preset proportion threshold such as 80, it is determined that it is a substantial block, triggering a second-level alarm, the current frame picture can be immediately intercepted and superimposed with a risk confidence and a heat map of a momentary spatial obstruction index, and pushed to a management cloud platform through a 4G / 5G network.

[0054] As shown in Figure 2 , it is a risk accumulation response mechanism schematic diagram in the embodiment of the application, it can be seen that whether it is a pedestrian or an article, the original signal captured by the sensor is maintained at a high level, and it is difficult to distinguish; and the risk accumulation confidence curve of the application is suppressed at a low level in a dynamic interference scene, and rapidly rises and breaks through the threshold in a static blocking scene, which embodies the technical effect of filtering dynamic false positives and locking static blocking of the application.

[0055] As shown in Figure 3 , it is a distribution model diagram of an emergency resource operation obstruction potential field, it can be seen that the heat color gradually becomes lighter from the center point outward, and once an object enters the center point area, the spatial obstruction potential strength will rapidly rise, which embodies the spatial perception ability of the system to protect key parts.

[0056] In this way, it can effectively distinguish between a slightly swinging human body and a completely stationary object, eliminate false positives caused by common scenes such as pedestrians leaning, and at the same time ensure the sensitive capture of real illegal stacking behaviors.

[0057] The embodiment of the application also discloses an emergency resource information warning system based on AI recognition, comprising a processor and a memory, the memory stores computer program instructions, and when the computer program instructions are executed by the processor, a kind of emergency resource information warning method based on AI recognition according to the application is realized.

[0058] The system also comprises other components well known to those skilled in the art such as a communication bus and a communication interface, the arrangement and function of which are known in the art and will therefore not be described in more detail here.

[0059] In the description of the present specification, the meaning of "a plurality of" or "several" is at least two, for example, two, three or more, and the like, unless explicitly specifically defined otherwise.

[0060] While the present specification has shown and described a number of embodiments of the application, it is to be understood that such embodiments are merely illustrative of the many possible embodiments thereof. Numerous modifications, adaptations, and variations will be apparent to those skilled in the art in view of the foregoing description. It is to be understood that, in the course of practicing the present application, embodiments of the application described herein can be adopted to employ each of the alternatives, modifications and variations.

Claims

1. A method for early warning of emergency medical resource information based on AI recognition, characterized in that, include: Establish a coordinate system for the monitoring scene, mark the center point of the operating handle of the emergency facility in the image as the singularity of the potential energy field, and divide the physical contour area of ​​the emergency cabinet as the region of interest. At the same time, collect the average ambient brightness value as the ambient lighting reference. The video stream is accessed in real time. The background is removed using a Gaussian mixture background modeling algorithm to extract the foreground moving target. The foreground moving target is morphologically processed to form a complete connected component. The real area ratio of the connected component is calculated. The foreground moving target is initially screened for attributes based on the real area ratio to determine the normalization adjustment coefficient. The geometric centroid of the connected component is calculated. Based on the straight-line distance from the geometric centroid to the center point of the operating handle, the normalization adjustment coefficient, and the pixel area of ​​the connected component, a preset logarithmic potential energy field model is used to calculate the instantaneous spatial obstacle index generated by the object in the current frame. The instantaneous spatial obstacle index is positively correlated with the normalization adjustment coefficient and the pixel area, and negatively correlated with the straight-line distance. The calculation of the instantaneous spatial obstacle index generated by the object in the current frame using the preset logarithmic potential energy field model includes: calculating the ratio of the potential energy radiation radius to the distance factor, adding 1 to the ratio, and taking the natural logarithm to obtain the distance attenuation factor. The distance factor is composed of the pixel distance from the object's centroid to the emergency cabinet handle and the safety buffer constant. The instantaneous spatial obstacle index is obtained by multiplying the normalization adjustment coefficient, the pixel area of ​​the foreground connected component, and the distance attenuation factor. The volatility change rate is calculated based on the instantaneous spatial obstacle index of the current frame and the previous frame. A stability weighting factor is constructed using the volatility change rate. The risk confidence level of the current moment is iteratively calculated by combining the risk confidence level accumulated at the previous moment and the natural decay constant. When the risk confidence level exceeds a preset threshold, a graded warning is executed. The calculation of the volatility change rate based on the instantaneous spatial obstacle index of the current frame and the previous frame includes: reading the instantaneous spatial obstacle index calculated in the previous frame and the instantaneous spatial obstacle index of the current frame; calculating the absolute value of the difference between the instantaneous spatial obstacle index of the current frame and the instantaneous spatial obstacle index of the previous frame as the volatility change rate. The iterative calculation of the risk confidence level of the current moment includes: calculating the stability weighting factor using an exponential function, where the exponential part of the stability weighting factor is the negative value of the product of the stability sensitivity coefficient and the volatility change rate; multiplying the instantaneous spatial obstacle index of the current frame by the stability weighting factor to obtain the risk increment of the current frame; adding the risk confidence level accumulated at the previous moment to the risk increment of the current frame and subtracting the natural decay constant to obtain the final risk confidence level accumulated at the current moment.

2. The method for early warning of emergency medical resource information based on AI recognition according to claim 1, characterized in that, The step of performing initial attribute screening of the foreground moving target based on the actual volume ratio to determine the normalization adjustment coefficient includes: Calculate the real area ratio of the connected component, where the real area ratio is the ratio of the area of ​​the connected component to the area of ​​its convex hull. If the actual area ratio is lower than a preset actual area ratio threshold, it is marked as a low-risk suspected target and the value of the normalization adjustment coefficient is lowered; if it is not lower than the preset actual area ratio threshold, it is marked as a high-risk suspected target and the value of the normalization adjustment coefficient is maintained or increased.

3. The method for early warning of emergency medical resource information based on AI recognition according to claim 1, characterized in that, The normalization adjustment coefficient is dynamically adjusted based on the total image resolution and the morphological regularity of the foreground moving target. The morphological regularity is characterized by the real volume ratio and is used to suppress the weight of irregular objects.

4. The method for early warning of emergency medical resource information based on AI recognition according to claim 1, characterized in that, The calculation of the risk confidence level also includes a boundary condition constraint step: If the calculated risk confidence score is less than zero, then the risk confidence score is forcibly assigned a value of zero; If the calculated risk confidence score is greater than the set saturation upper limit, then the risk confidence score will be forcibly assigned the value of the saturation upper limit.

5. The method for early warning of emergency medical resource information based on AI recognition according to claim 1, characterized in that, The implementation of tiered early warning includes: When the risk confidence level exceeds the first preset ratio threshold and continues to rise, a level one warning is triggered and a voice prompt is played. When the risk confidence level is greater than the second preset ratio threshold, it is determined to be a substantial obstruction, triggering a level two alarm. The current frame is captured and a heat map of the risk confidence level and the instantaneous spatial obstacle index is superimposed and pushed to the management cloud platform.

6. The method for early warning of emergency medical resource information based on AI recognition according to claim 1, characterized in that, The straight-line distance is Euclidean distance, and the center point of the operating handle is the singularity of the potential energy field, which is the area with the highest protection level.

7. An emergency medical resource information early warning system based on AI recognition, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement an AI-based emergency resource information early warning method according to any one of claims 1-6.

Images