Anti-drowning early warning system of self-adaptive dynamic virtual electronic fence

Abstract

The invention discloses an anti-drowning early warning system of a self-adaptive dynamic virtual electronic fence, and aims to solve the problems that an existing anti-drowning early warning system cannot adapt to water level changes and early warning lags behind. The virtual electronic fence setting module is used for setting static and dynamic virtual electronic fences, identifying and analyzing personnel behaviors and controlling alarms; the data acquisition module is used for acquiring environment data information and video monitoring information; and the data storage module is used for storing data, alarm information, virtual electronic fence setting information and water level change information of the virtual electronic fence according to the water level change adjustment range. The system is based on real-time detection of a YOLOv8 model and is combined with a virtual electronic fence technology, has the characteristics of high-precision identification, low-delay response, field environment adaptability and the like, can effectively reduce the occurrence rate of drowning accidents, and is a set of intelligent, efficient and convenient-to-maintain system.

Description

Technical Field

[0001] The present invention relates to an open water safety protection system, specifically a drowning prevention early warning system with an adaptive dynamic virtual electronic fence, and relates to the field of water level monitoring and dangerous area protection in environments such as ponds and lakes. Background Art

[0002] Open waters such as ponds and lakes are high-incidence areas of drowning accidents, especially for children and the elderly. Traditional water area safety protection measures, such as physical fences and warning signs, have high costs for physical fences, need to be installed along the water edge, are difficult to construct and maintain, and cannot dynamically adjust the safety boundary according to water level changes; moreover, the passive early warning ability of traditional protection measures is poor, and existing camera systems usually only support voice shouting, lacking the ability to actively identify intrusion behaviors and trigger linkage alarms; at the same time, the adaptability is insufficient, and static monitoring cannot cope with environmental changes such as water level fluctuations and vegetation occlusion, and it is easy to generate monitoring blind spots. Summary of the Invention

[0003] The present invention aims to provide a drowning prevention early warning system with an adaptive dynamic virtual electronic fence, which realizes real-time monitoring, dangerous behavior recognition and early warning of water areas through the combination of dynamic and static virtual electronic fences, reduces the incidence of drowning accidents, and realizes accurate drowning prevention early warning. To achieve the above object, the present invention provides the following technical solutions:

[0004] Preferably, a drowning prevention early warning system with an adaptive dynamic virtual electronic fence includes: a data acquisition module, a virtual electronic fence setting module; a data storage module; the data acquisition module includes: a video acquisition module and a water level monitoring module; the virtual electronic fence setting module includes: a static virtual electronic fence module, a dynamic virtual electronic fence module, a behavior recognition and analysis module, and a control alarm module; the data storage module includes: an alarm information storage module, a water surface - virtual electronic fence information storage module.

[0005] Preferably, for the video acquisition module, a high-definition camera network is deployed around the water area. The high-definition camera supports H.265 hardware encoding and wide dynamic range imaging. The imaging of the high-definition camera uses a hierarchical transmission architecture to push the video stream to the behavior recognition and analysis module through wired Gigabit Ethernet and wireless Wi-Fi6 Mesh, and an adaptive bitrate control algorithm is used to dynamically adjust the bitrate. The bitrate calculation formula is: where, R(t) is the real-time bitrate in Mbps, R base = 4 Mbps, B available (t) is the current available bandwidth (Mbps), B threshold = 2 Mbps.

[0006] More preferably, the water level monitoring module includes wireless water pressure sensors array - deployed at the bottom of the pond, in the middle of the pond, and 0.5 m above the pond. It is connected to the dynamic virtual electronic fence module through the LoRaWAN protocol, and real - time collects the pressure data of the wireless water pressure sensors and converts it into water level height based on the hydrostatic formula. The water level height conversion formula is: where h is the water level height (m), P is the measured pressure of the sensor (Pa), P 0 is the standard atmospheric pressure, and ρ(T) is the water density obtained by looking up the table according to the water temperature T (g = 9.8065 m / );

[0007] More preferably, the static virtual electronic fence module has a human - machine interaction interface. Managers can preset the basic safety boundary according to the water area terrain data and mark fixed dangerous areas such as deep pits and steep slopes.

[0008] More preferably, the dynamic virtual electronic fence module is based on real - time water level data. It predicts the water level change trend in the next half hour through the Gaussian process regression model, and uses the polygon geometric algorithm to generate an adaptive fence according to the water level change, dynamically adjusting the fence boundary. The Gaussian regression model is: where m(x) is the mean function, usually assumed to be zero, k(x,x') is the covariance function, X is the given training data, y is the corresponding observed value, x is the new input point, K is the covariance matrix of the training data, I is the identity matrix, with the same dimension as K, k is the covariance vector between the training data and the new input point, and σ 2 n is the noise variance. The polygon geometric algorithm is: where h(t) is the starting water level, P(t) is the starting fence boundary, ∆y i is the offset relative to the reference water level, h'(t) is the new water level prediction, and p'(t) is the new fence boundary.

[0009] More preferably, the behavior recognition and analysis module includes the following steps, S1 Use the YOLOv8 object detection algorithm to detect people in the real - time scene, calculate the distance from the static virtual electronic fence, perform dynamic intention recognition, and when it is judged that the distance between the person and the fence is less than the threshold, immediately trigger a pre - alarm; The YOLOv8 object detection algorithm is: S11 Bounding box representation and person center coordinate representation where (x min , y min ) and (x max , y max ) are the coordinates of the upper left corner and the lower right corner of the bounding box respectively, (x c , y c ) is the coordinate of the person center point, and Box is the detected person bounding box; S12 Calculate the shortest distance from the person center point to the boundary of the electronic fence Assume that the electronic fence is represented by a polygon, and its boundary consists of a series of vertices {(x 1 , y 1 ), (x 2 , y 2 ) ⋯, (x n , y n )}; Traverse all fence edges to find the shortest distance; S13 Calculate the velocity vector v = (v x , v y ) according to the position change of the person center point in consecutive frames S14 Calculate the direction angle θ S15 Direction judgment: Determine whether it points to the fence by the angle between the velocity vector and the vector d from the person to the fence: When cosφ > 0, the movement direction points to the fence; S16 Early warning trigger: When the minimum distance from the person to the fence is less than the preset threshold and the movement direction points to the fence, trigger an early warning; S2 Use the target tracking algorithm to judge the person's crossing behavior. If there is a crossing, it is determined as a valid alarm; The target tracking algorithm is as follows: S21 Kalman filter is used to predict the position of the person in the next frame: where x t is the state vector including the predicted position and velocity of the person, A is the state transition matrix, B is the control matrix, u t is the control input, and w t is the process noise; S22 Crossing judgment: When the position of the person crosses the fence boundary, it is determined as a crossing; S23 Crossing determination: If the person enters from the outside to the inside, it is determined as a valid crossing alarm; Use the ray intersection method to determine whether the center point of the person is inside the polygonal fence: Shoot a horizontal ray to the right from the center of the person; calculate the number of intersections between the ray and the fence boundary; If the number of intersections is odd, the point is inside the polygon, otherwise it is outside.

[0010] Preferably, the control alarm module adopts a graded alarm setting, including pre-alarm and out-of-bounds alarm; specifically, pre-alarm: voice broadcast to remind people to stay away from dangerous areas, out-of-bounds alarm: trigger the sound and light alarm, and simultaneously send positioning information to management personnel and rescue personnel, so as to detect when people enter the dangerous area or there is a potential risk, and trigger the corresponding level of alarm.

[0011] More preferably, the data storage module and the water surface-virtual electronic fence information storage module save water level change data and dynamic electronic fence adjustment records, and have data backup and recovery functions to ensure that the fence range is consistent with water level changes, assist the system in analyzing potential dangerous areas in the water area, and adjust the position of the electronic fence in time to adapt to real-time changes in the water level and predict the water level change trend in the future; the alarm information storage module records historical alarm data to determine the possibility of drowning of people approaching the water area.

[0012] The beneficial effects of the present invention are:

[0013] Pond drowning prevention is an important part of drowning prevention work. However, most of the current protection is only through the establishment of simple warning signs such as "Danger Water Depth", and there is still a lack of necessary drowning prevention systems.

[0014] The virtual electronic fence setting module of this system can significantly improve the ability to judge dangerous behaviors, enhance the all-day climate perception, and realize the behavior prediction mechanism through the multi-target trajectory tracking algorithm, thereby realizing real-time identification of dangerous behaviors such as people falling into the water and crossing the boundary, and can still maintain high reliability and high applicability at night or in complex weather (such as heavy rain and fog). The data storage module builds an intelligent early warning system through hierarchical early warning and response efficiency optimization to achieve full-link closed-loop management from risk warning to emergency rescue. When a dangerous event occurs, the system uses a multi-channel hierarchical response mechanism, pre-alarms and cross-border alarms to effectively reduce the incidence of drowning accidents. The data acquisition module pioneered the "water level-weather-fence" linkage model, and at the same time achieved full coverage of dangerous areas through dynamic virtual fences, realizing effective monitoring of dangerous areas.

[0015] In summary, with the core of "precision detection - dynamic protection - rapid response - continuous optimization", this system provides an efficient, economical and scalable anti-drowning protection for open waters through technology integration and data-driven approach. It can effectively reduce the drowning risk. Through technology optimization, this system not only reduces the renovation and maintenance costs, simplifies the operation management process, significantly improves the work efficiency, but also greatly reduces the human resource requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a structural diagram of a pond anti-drowning warning system with an adaptive dynamic virtual electronic fence according to the present invention.

[0017] Figure 2 is a module structural diagram of a pond anti-drowning warning system with an adaptive dynamic virtual electronic fence according to the present invention.

[0018] Figure 3 is an operation flow chart of a pond anti-drowning warning system with an adaptive dynamic virtual electronic fence according to the present invention.

[0019] Figure 4 is a schematic diagram of the virtual electronic fence area of a pond anti-drowning warning system with an adaptive dynamic virtual electronic fence according to the present invention.

[0020] Figure 5 is a three-dimensional deployment profile schematic diagram of a wireless water pressure sensor of a pond anti-drowning warning system with an adaptive dynamic virtual electronic fence according to the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] Aiming at the long-existing technical shortcomings in the domestic pond anti-drowning management field (such as insufficient monitoring coverage, lagged response, low efficiency of manual inspection, etc.), the present invention provides an anti-drowning warning system with an adaptive dynamic virtual electronic fence, which realizes real-time monitoring of water area safety through the YOLOv8 algorithm. This virtual electronic fence can conduct preliminary monitoring around the pond, and judge potential conflict risks by detecting whether there are pedestrians falling into the pond and the risk of pedestrians falling into the pond.

[0022] An anti-drowning warning system with an adaptive dynamic virtual electronic fence, as shown in the attached Figure 2 figures, includes: a virtual electronic fence setting module, which is used for setting the virtual electronic fence, judging the potential risks of personnel targets approaching dangerous areas and subsequent timely rescue; a data acquisition module, which is used for real-time detecting personnel targets around the pond and the change of the pond water surface; a data storage module, which is used for storing data on the range adjusted by the virtual electronic fence according to water level changes, alarm information, virtual electronic fence setting information and water level change information.

[0023] 1. The data acquisition module includes a video acquisition module and a water level monitoring module;

[0024] The described video acquisition module deploys a high-definition camera network around the water area. The high-definition cameras support H.265 hardware encoding and wide dynamic range imaging. The imaging of the high-definition cameras uses a hierarchical transmission architecture to push the video stream to the behavior recognition and analysis module through wired Gigabit Ethernet and wireless Wi-Fi6 Mesh, and adopts an adaptive bitrate control algorithm to dynamically adjust the bitrate. The bitrate calculation formula is: where, R(t) is the real-time bitrate in Mbps, R base = 4 Mbps, B available (t) is the current available bandwidth (Mbps), B threshold = 2 Mbps.

[0025] The described water level monitoring module is deployed at the bottom of the pond through a wireless water pressure sensor 6, a sensor 5 is deployed in the middle of the pond, and a sensor 4 is deployed 0.5 meters above the pond; it is connected to the dynamic virtual electronic fence module through the LoRaWAN protocol, and real-time collects the pressure data of the wireless water pressure sensor and converts it into the water level height based on the hydrostatic formula; the water level height conversion formula is: where, h is the water level height (m), P is the measured pressure of the sensor (Pa), P 0 is the standard atmospheric pressure, ρ(T) is the water density obtained by looking up the table according to the water temperature T (g = 9.8065 m / s 2 );

[0026] 2. The virtual electronic fence setting module includes a static virtual electronic fence module, a dynamic virtual electronic fence module, a behavior recognition and analysis module, and a control alarm module;

[0027] The described static virtual electronic fence module has a human-computer interaction interface. Managers can preset the basic safety boundary according to the water area terrain data and mark fixed dangerous areas such as deep pits and steep slopes.

[0028] The described dynamic virtual electronic fence module, based on real-time water level data, predicts the future half-hour water level change trend through a Gaussian process regression model, and uses a polygon geometric algorithm to generate an adaptive fence according to the water level change and dynamically adjust the fence boundary; The Gaussian regression model is: where, m(x) is the mean function, usually assumed to be zero, k(x,x') is the covariance function, X is the given training data, y is the corresponding observed value, x For a new input point, K is the covariance matrix of the training data, I is the identity matrix with the same dimension as K, and k is the covariance vector between the training data and the new input point, and σ 2 n is the noise variance. The polygon geometric algorithm is as follows: where h(t) is the initial water level, P(t) is the initial fence boundary, ∆y i is the offset relative to the reference water level, h'(t) is the new water level prediction, and p'(t) is the new fence boundary

[0029] The behavior recognition and analysis module includes the following steps S1 Use the YOLOv8 object detection algorithm to detect people in the real-time scene, calculate the distance from the static virtual electronic fence, perform dynamic intention recognition, and immediately trigger a pre-alarm when the distance between the person and the fence is less than the threshold; The YOLOv8 object detection algorithm is as follows: S11 Bounding box representation and person center coordinate representation where (x min , y min ) and (x max , y max ) are the upper-left and lower-right coordinates of the bounding box respectively, (x c , y c ) is the person center point coordinate, and Box is the detected person bounding box; S12 Calculate the shortest distance from the person center point to the electronic fence boundary Assume that the electronic fence is represented by a polygon, and its boundary consists of a series of vertices {(x 1 , y 1 ), (x 2 , y 2 ) ⋯, (x n , y n )} Traverse all fence sides to find the shortest distance; S13 Calculate the velocity vector v = (v x , v y ) according to the position change of the person center point in consecutive frames S14 Calculate the direction angle θ S15 Direction judgment: Determine whether it points to the fence by the angle between the velocity vector and the vector d from the person to the fence: When cosφ > 0, the moving direction points to the fence; S16 Early warning trigger: When the minimum distance between the person and the fence is less than the preset threshold and the moving direction points to the fence, an early warning is triggered; S2 Uses a target tracking algorithm to judge the behavior of a person crossing the boundary. If there is a boundary crossing, it is determined as a valid alarm; The target tracking algorithm is as follows: S21 Kalman filtering is used to predict the next frame position of the person: Among them, x t is the state vector including the predicted position and speed of the person, A is the state transition matrix, B is the control matrix, u t is the control input, w t is the process noise; S22 Boundary crossing judgment: When the position of the person crosses the fence boundary, it is determined as a boundary crossing; S23 Boundary crossing determination: If the person enters from the outside to the inside, it is determined as a valid boundary crossing alarm; Use the ray intersection method to judge whether the center point of the person is inside the polygon fence: Emit a ray horizontally to the right from the center point of the person; Calculate the number of intersection points of the ray and the fence boundary; If the number of intersection points is odd, the point is inside the polygon, otherwise it is outside.

[0030] The described control alarm module adopts a hierarchical alarm setting, including an early warning and a boundary crossing alarm; specifically, the early warning: voice broadcast to remind the person to stay away from the dangerous area, the boundary crossing alarm: trigger an audible and visual alarm, and synchronously send the positioning information to the management personnel and rescue personnel, so as to trigger the corresponding level of alarm when it is detected that the person enters the dangerous area or there is a potential risk.

[0031] Among them, when the target person approaches the static virtual electronic fence:

[0032] a. According to the relevant parameters of the pond previously entered into the system, the corresponding static virtual electronic fence has been generated.

[0033] b. When the driving route of the target person is close to the virtual electronic fence by 2m and continues to approach, the system will recognize the potential risk and trigger an early warning in a timely manner; when the target person crosses the virtual electronic fence, the early warning alarm will be triggered, and the positioning information of the target person will be accurately transmitted to the administrator and rescue personnel through the Beidou positioning system.

[0034] 3. The data storage module includes a water surface - virtual electronic fence information storage module and an alarm information storage module;

[0035] The described water surface - virtual electronic fence information storage module saves water level change data and dynamic electronic fence adjustment records, with data backup and recovery functions, ensuring that the fence range is consistent with the water level change, assisting the system in analyzing potential dangerous areas in the water area, and timely adjusting the position of the electronic fence to adapt to the real-time change of the water level and predicting the water level change trend in a future period of time;

[0036] The described alarm information storage module records historical alarm data and judges the drowning possibility of personnel approaching the water area.

[0037] The following combines the attached Figure 3 and the attached Figure 4 to illustrate the dynamic virtual electronic fence setting and personnel recognition process.

[0038] When the sensor detects that the change in the water surface exceeds the threshold of the static virtual electronic fence 1, the terminal device 2 will predict the water level change trend in the next half hour according to the real-time water level data, and generate an adaptive fence based on the water level change using the polygon geometric algorithm, dynamically adjusting the fence boundary;

[0039] When the target person enters the monitoring range, the YOLOv8 detection algorithm is used to detect the personnel in the real-time scene, calculate the distance from the static virtual electronic fence, perform dynamic intention recognition, and immediately trigger a pre-alarm 3 when it is judged that the distance between the person and the fence is less than the threshold. The target tracking algorithm is used to judge the overstepping behavior of the person. If there is overstepping, it is determined as a valid alarm.

Claims

1. An adaptive dynamic virtual electronic fence drowning prevention warning system, comprising: Data acquisition module, virtual electronic fence setting module, data storage module; The data acquisition module includes: a video acquisition module and a water level monitoring module; the virtual electronic fence setting module includes: a static virtual electronic fence module, a dynamic virtual electronic fence module, a behavior recognition and analysis module and a control alarm module; the data storage module includes: an alarm information storage module and a water surface-virtual electronic fence information storage module.

2. The video acquisition module according to claim 1, characterized in that: A high-definition camera network is deployed around the water area. The high-definition camera supports H.265 hardware encoding and wide dynamic range imaging. The high-definition camera imaging uses a layered transmission architecture to push the video stream to the behavior recognition and analysis module using wired Gigabit Ethernet and wireless Wi-Fi6 Mesh, and uses an adaptive bit rate control algorithm to dynamically adjust the bit rate. The bit rate calculation formula is: in, R ( t ) is the real-time bit rate in Mbps, R base =4 Mbps, B available ( t ) is the current available bandwidth in Mbps, B threshold = 2 Mbps.

3. The water level monitoring module according to claim 1, characterized in that: Wireless water pressure sensors are deployed in arrays at the bottom of the pond, in the middle of the pond, and 0.5 meters above the pond. They are connected to the dynamic virtual electronic fence module through the LoRaWAN protocol to collect pressure data from the wireless water pressure sensors in real time and convert it into water level height based on the fluid statics formula. The water level height conversion formula is: in, h is the water level height (m), P is the actual pressure measured by the sensor (Pa), P 0 is standard atmospheric pressure, ρ(T) is the water density (g=9.8065m / s 2 ).

4. The static virtual electronic fence module according to claim 1, characterized in that: The static virtual electronic fence module has a human-computer interactive interface, and managers can preset basic safety boundaries according to water terrain data, and mark fixed dangerous areas such as deep pits and steep slopes.

5. The dynamic virtual electronic fence module according to claim 1, characterized in that: The dynamic virtual electronic fence module is based on real-time water level data, uses a Gaussian process regression model to predict the water level change trend in the next half hour, and uses a polygonal geometry algorithm to generate an adaptive fence according to the water level change, and dynamically adjusts the fence boundary; The Gaussian regression model is: in, m ( x ) is the mean function, which is usually assumed to be zero. k ( x , x´ ) is the covariance function, X For a given training data, y is the corresponding observed value, x * is the new input point, K is the covariance matrix of the training data, I is the identity matrix, with the same dimensions as K same, k * is the covariance vector between the training data and the new input point, σ 2 n is the noise variance. The polygon geometry algorithm is: Where h(t) is the initial water level, P(t) is the initial fence boundary, ∆y i is the offset relative to the reference water level, h′(t) is the new water level prediction, and p′(t) is the new fence boundary.

6. The behavior recognition and analysis module according to claim 1, characterized in that: The following steps are included: S1 uses the YOLOv8 target detection algorithm to detect people in real-time scenes, calculate the distance to the static virtual electronic fence, perform dynamic intention recognition, and immediately trigger a pre-alarm when it determines that the distance between the person and the fence is less than the threshold; The YOLOv8 target detection algorithm is: S11 bounding box representation and person center coordinate representation Among them, (x min ,y min ) and (x max ,y max ) are the coordinates of the upper left corner and lower right corner of the bounding box, respectively, (x c ,y c ) is the coordinate of the center point of the person, Box is the detected person boundary box; S12 calculates the shortest distance from the center point of the person to the boundary of the electronic fence Assume that the geo-fence is represented by a polygon whose boundary is a series of vertices {(x1,y1),(x2,y2)⋯,(x n ,y n )}composition Traverse all fence edges and find the shortest distance; S13 calculates the velocity vector v=(v x ,v y ) S14 calculates the direction angle θ S15 direction judgment: judge whether it is pointing to the fence by the angle between the speed vector and the vector d from the person to the fence: When cosφ>0, the direction of movement is toward the fence; S16 Pre-alarm trigger: When the minimum distance between a person and the fence is less than the preset threshold and the direction of movement is toward the fence, a pre-alarm is triggered; S2 uses a target tracking algorithm to determine whether a person has crossed the boundary. If there is a crossing of the boundary, it is considered a valid alarm. The target tracking algorithm is: The S21 Kalman filter is used to predict the next frame position of the person: in, x t is the state vector including the predicted position and speed of the personnel, A is the state transfer matrix, B is the control matrix, u t is the control input, w t is the process noise; S22: when a person's position crosses the fence boundary, it is judged as crossing the boundary; S23 cross-border judgment: if a person enters from the outside, it is judged as a valid cross-border alarm; Use the ray intersection method to determine whether the center point of the person is inside the polygonal fence: Send a horizontal ray to the right from the center point of the person; Count the number of intersections between the ray and the fence boundary; If the number of intersections is odd, the point is inside the polygon, otherwise it is outside.

7. The control alarm module according to claim 1, characterized in that: A graded alarm setting is adopted, including pre-alarm and out-of-bounds alarm; specifically, pre-alarm: voice broadcast to remind people to stay away from dangerous areas, out-of-bounds alarm: trigger the sound and light alarm, and simultaneously send positioning information to management personnel and rescue personnel, so as to trigger the corresponding level of alarm when it is detected that people have entered the dangerous area or there is a potential risk.

8. The data storage module according to claim 1, characterized in that: The water surface-virtual electronic fence information storage module saves water level change data and dynamic electronic fence adjustment records, and has data backup and recovery functions to ensure that the fence range is consistent with water level changes. It assists the system in analyzing potential dangerous areas in the water area and adjusts the position of the electronic fence in time to adapt to real-time changes in the water level and predict the water level change trend in the future; the alarm information storage module records historical alarm data to determine the possibility of drowning of people approaching the water area.

Images