Swimming pool drowning prevention early warning system and method based on artificial intelligence

Abstract

The invention belongs to the field of swimming pool safety monitoring, and particularly relates to a swimming pool drowning prevention early warning system and method based on artificial intelligence. The swimming pool drowning prevention early warning system based on artificial intelligence comprises an acquisition module used for sending basic questionnaire data including personal information, swimming experience, water activity preference and body characteristics to a user terminal, and speculating suffocation time and training frequency of a user according to a filling result of a basic questionnaire; the acquisition module comprises a plurality of cameras, and the plurality of cameras are used for acquiring video data at any position of the water surface; the recognition module comprises a preset recognition model, and the recognition model is used for obtaining the video data, obtaining the head features and the water surface features in the video data through the recognition model, and obtaining the water entering time and the water exiting time of the head according to the relative position relation between the features of the five sense organs in the head features and the water surface features. According to the scheme, the problem that the accuracy rate is low when video monitoring is used for judging whether the swimmer is drowning or not is solved.

Description

Technical Field

[0001] This solution belongs to the field of swimming pool safety monitoring, and specifically involves an artificial intelligence-based swimming pool drowning prevention warning system and method. Background Art

[0002] With the development of sports, people's enthusiasm for participating in sports activities has also increased. However, swimming, as one of the most popular sports, has become the sport with the highest incidence of safety accidents. According to incomplete statistics released by the Ministry of Health, approximately 57,000 people die from drowning accidents in my country each year, of which 56.04% are young people, making it the leading cause of death among young people. China also has the highest drowning death rate in the world.

[0003] Existing methods for identifying drowning in swimming pools primarily include lifeguard monitoring, wearable monitoring, and video monitoring. Lifeguard monitoring is the most traditional monitoring method. During the monitoring process, lifeguards must endure prolonged nervous tension and fatigue, resulting in low monitoring efficiency for drowning incidents, which in turn affects timely rescue of drowning swimmers. While wearable monitoring solves the problem of low monitoring efficiency in traditional drowning prevention monitoring, the comfort of wearing wearable devices while swimming is poor. Prolonged wear can cause pressure on the wrist, reducing the swimmer's swimming experience. Consequently, many swimmers remove their wearable devices mid-swim, resulting in poor monitoring effectiveness. Video monitoring solves the problem of low monitoring efficiency in traditional drowning prevention monitoring without affecting the swimmer's swimming experience, making it an excellent method for identifying drowning in swimming pools. However, when existing video surveillance performs image recognition on people drowning in swimming pools, it usually uses a fixed breath-holding time or paddling action to distinguish whether the swimmer is drowning. However, the breath-holding time of different swimmers varies greatly, and the movements of drowning struggles and playful splashing water are relatively similar, making video surveillance less accurate in determining whether a swimmer is drowning. Summary of the Invention

[0004] The purpose of this solution is to provide an artificial intelligence-based swimming pool drowning prevention warning system and method to solve the problem of low accuracy of video surveillance in determining whether a swimmer is drowning.

[0005] To achieve the above objectives, this solution provides an artificial intelligence-based swimming pool drowning prevention warning system, which is characterized by including: A collection module is configured to send a basic questionnaire to the user terminal, which contains personal information, swimming experience, water activity preferences, and physical characteristics, and to estimate the user's breath-holding time and training frequency based on the results of the basic questionnaire. The collection module includes multiple cameras, which are configured to collect video data from any position on the water surface. A recognition module, including a preset recognition model, is used to acquire video data and, using the recognition model, to obtain head features and water surface features in the video data, and to obtain the entry and exit times of the head based on the relative positional relationship between the facial features of the head and the water surface features. The recognition module is also used to match personal information based on the head features to obtain a corresponding breath-holding time and training frequency range. The recognition module is used to count according to the breath-holding time when the entry time is obtained, and if the exit time is not obtained at the end of the timing, the video data and life-saving prompts are sent to the lifeguard terminal; The recognition module is further configured to count the number of water immersion times obtained during the breath-holding time according to the personal information as the number of water immersions, obtain the water immersion frequency according to the number of water immersions, and send video data and lifesaving prompts to the lifeguard terminal if the water immersion frequency exceeds the training frequency range; The recognition module is also used to obtain the interval between two adjacent water immersion times when the water immersion frequency is within the training frequency range. If the interval time difference is greater than a preset difference, the video data and lifesaving prompts are sent to the lifeguard terminal.

[0006] And, an artificial intelligence-based swimming pool drowning prevention warning method using an artificial intelligence-based swimming pool drowning prevention warning system.

[0007] The principle and technical effect of this solution are as follows: First, this solution first collects basic questionnaire data containing personal information, swimming experience and other aspects through the acquisition module to infer key information such as the user's breath-holding time and training frequency. Then, the recognition module uses the relative position relationship between the head features and the water surface features in the video data to derive the entry time, exit time, etc., and then combines the relevant parameters inferred from the previous questionnaire to make a comprehensive judgment. This comprehensive consideration can more accurately distinguish between normal and drowning situations, effectively reduce misjudgments, and improve accuracy. In addition, the recognition module not only focuses on single entry and exit situations, but also deeply analyzes the entry frequency and the interval time between two adjacent entries. This detailed behavioral analysis can more keenly capture potential drowning risks. Compared with traditional video surveillance that only generally checks whether there is struggling in the water, it greatly improves the accuracy of judging drowning situations.

[0008] Secondly, head features not only indicate a swimmer's drowning condition but also quickly identify their personal information through facial features, allowing lifeguards to quickly locate potential drowning individuals in a crowd. Furthermore, this personal information, combined with the swimmer's individual swimming level, can be used to determine a more precise breath-holding time and training frequency range. This solution relaxes some of the criteria for normal swimming behavior to avoid frequent false alarms. While ensuring the safety of swimmers of varying levels, it also prevents overly sensitive alarms from disrupting normal swimming activities, making pool monitoring more intelligent and user-friendly.

[0009] In summary, this solution solves the problem of low accuracy of video surveillance in determining whether a swimmer is drowning.

[0010] Furthermore, the recognition model includes a target detection model established by Faster R-CNN, SSD or YOLO, and the recognition module uses a face dataset containing the positions of key points of facial features and a dataset containing various postures of the head to train the target detection model; the recognition model also includes a water surface segmentation model established by U-Net or DeepLab, and the recognition model uses the water surface segmentation model to identify and segment the water surface in the video data, and then uses the gray-level co-occurrence matrix to extract the texture features of the water surface, and finally obtains the water surface features.

[0011] Furthermore, the recognition model is also used to identify the behavior information of the swimmer. The recognition module calculates the optical flow through the Lucas-Kanade optical flow algorithm or the FlowNet method based on deep learning, analyzes the motion trajectory of the human head according to the calculated optical flow information, obtains the swimmer's water exit and water entry behaviors according to the motion trajectory of the human head, and records the time corresponding to the water exit and water entry behaviors as the water exit time and water entry time respectively; the recognition module also includes a human posture estimation model constructed by OpenPose, which infers the posture of the human body by identifying the positions of the key points of the human body, such as the neck, shoulders, and torso nodes. The recognition model combines continuous posture estimation results into an action sequence, uses a recurrent neural network or a long short-term memory network model for action recognition, and judges the swimmer's continuous posture as action behavior by analyzing the action sequence. The action behavior includes swimming action, playing action and resting action, and sets corresponding training frequency ranges for different action behaviors of the swimmer.

[0012] Furthermore, when the recognition module matches personal information based on head features to obtain the corresponding breath-holding time and training frequency range, it first uses the target detection model to identify the swimmer's facial features, and matches the swimmer's personal information based on the facial features; at the same time, it uses the human posture estimation model to infer the swimmer's body posture, and then obtains the swimmer's movement behavior based on the swimmer's continuous body posture; the swimmer's movement behavior and personal information are combined to obtain the swimmer's current breath-holding time and training frequency range.

[0013] By identifying and differentiating swimmers' behaviors and adjusting their breath-holding duration and training frequency range, the system achieves personalized drowning judgment criteria. Different swimmers have different abilities and habits. This allows the system to better adapt to individual differences, reducing the chances of inaccurate judgments based on a single standard across swimmers of varying abilities, further improving the accuracy of drowning judgments.

[0014] Furthermore, the recognition module is used to obtain the number of other swimmers within the preset range of the swimmer when a play action is recognized. If the number is less than 1, the swimmer is regarded as a dangerous swimmer, and the dangerous swimmer's action behavior, interval time, breath holding time and water entry frequency are continuously monitored until other swimmers appear within the preset range of the dangerous swimmer; the recognition module is used to send video data corresponding to the dangerous swimmer to the lifeguard terminal when a dangerous swimmer is recognized.

[0015] This approach takes into account potential unexpected situations during play, as swimmers' attention is often distracted and unaccompanied, increasing the potential risk. By prioritizing these swimmers, we can monitor dangerous situations (such as drowning) at an early stage. Compared to focusing solely on signs of drowning, this approach can detect anomalies earlier, reducing the likelihood of missed detections and improving the accuracy of overall drowning assessments. Furthermore, sudden cessation of play, unusually long intervals between swims, or a sudden increase in the frequency of water immersion during play can all be precursors to drowning. This continuous monitoring mechanism helps more accurately determine whether a swimmer is in danger in complex play scenarios, avoiding misjudgments caused by missing fleeting behavioral changes. Furthermore, this approach helps pool managers more precisely allocate safety resources. Based on the distribution of dangerous swimmers identified by this solution, lifeguards can be more effectively positioned and prioritized, avoiding wasted resources. Furthermore, by proactively identifying potentially dangerous situations, pool managers (lifeguards or other non-lifeguarding personnel) can take preventative measures. This preventative management approach helps reduce the incidence of drowning accidents and improve overall pool safety.

[0016] Furthermore, the lifeguard terminal includes a prompt playback terminal and a video display terminal. The video display terminal is used to play the received video data. The video display terminal stores a venue simulation map. The video display terminal is also used to obtain the collection position of the video data and mark a warning red dot at a position corresponding to the collection position in the venue simulation map. The video display terminal is used to display the venue simulation map; the prompt playback terminal is used to play a life-saving prompt when a life-saving signal is received, and highlight the warning red dot corresponding to the life-saving signal in the venue simulation map displayed on the video display terminal.

[0017] Furthermore, the lifeguard terminal includes several prompt playback terminals, the prompt playback terminal stores the lifeguard number, and the prompt playback terminal is provided with a confirmation button; the prompt playback terminal is used to use the information entered by the confirmation button as the lifesaving confirmation information when playing the lifesaving prompt, and send the lifesaving confirmation information, the lifeguard number and the lifesaving prompt to the video display terminal, and the video display terminal stops flashing the warning red dot corresponding to the lifesaving prompt after receiving the confirmation information, and counts down the preset first arrival time; if the lifeguard corresponding to the lifeguard number is collected in the video data corresponding to the lifesaving prompt before the countdown ends, the important lifesaving prompt and the video data corresponding to the important lifesaving prompt are stopped from playing, and the important lifesaving prompt is deleted from the venue simulation map. The warning red dot corresponding to the lifesaving prompt stops the countdown; if after the countdown of the first arrival time ends, the lifeguard corresponding to the lifesaving prompt is not collected in the video data corresponding to the lifesaving prompt. Then, an important lifesaving prompt is sent to the prompt playback terminal according to the lifesaving prompt; after the video display terminal sends the important lifesaving prompt, the warning red dot corresponding to the important lifesaving prompt flashes at a preset frequency; after receiving the important lifesaving prompt, the prompt playback terminal plays the important lifesaving prompt; until the appearance features of the lifeguard are collected in the video data corresponding to the important lifesaving prompt, the playback of the important lifesaving prompt and the video data corresponding to the important lifesaving prompt is stopped, and the warning red dot corresponding to the important lifesaving prompt is deleted from the venue simulation map.

[0018] By linking the prompt playback terminal with the video display terminal, lifeguards are prompted to promptly confirm and respond to lifesaving prompts. This mechanism ensures that lifeguards can act quickly upon receiving information about a possible drowning event and allows real-time tracking of their arrival at the scene. If the lifeguard does not arrive at the location corresponding to the video data within the specified time (the first arrival time), a critical lifesaving prompt is sent and the corresponding video data is highlighted. This helps to promptly identify potential problems in the lifesaving process, indirectly improving the timeliness and effectiveness of overall drowning rescue. This also helps confirm or correct the solution's initial judgment of whether a swimmer is drowning through the lifeguard's timely feedback and actions.

[0019] Furthermore, the recognition module is also used to extract the swimmer's action behavior and pool position from historically collected video data based on the swimmer's personal information, and use the extracted action behavior and pool position to establish a probability model; the recognition module is used to obtain the swimmer's current action behavior and current pool position from the currently received video data, and put the current action behavior and current pool position into the probability model to infer the rationality of the swimmer's current behavior. When the rationality is lower than a preset rational value, the swimmer is regarded as a dangerous swimmer.

[0020] This solution extracts swimmers' movements and pool positions from historically collected video data to establish a probabilistic model. This provides a benchmark based on past experience for determining whether a swimmer's behavior is reasonable. This probabilistic model accurately captures deviations from typical patterns, promptly identifying irrational behavior and determining potential drowning risks. This effectively reduces the number of anomalies missed by relying solely on real-time imagery and improves the accuracy of drowning judgments. Furthermore, this mechanism for identifying dangerous swimmers in advance encourages staff to provide appropriate guidance and attention to them, shifting from passive rescue to active prevention, and ultimately reducing the likelihood of drowning accidents. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Schematic diagram of the structure of an artificial intelligence-based swimming pool drowning prevention warning system in an embodiment of the present invention.

[0022] Figure 2 This is a flowchart of sending video data and life-saving tips to a lifeguard terminal based on the breath-holding time in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The following will clearly and completely describe the concept and technical effects of the present invention in conjunction with the embodiments to fully understand the purpose, features and effects of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, other embodiments obtained by those skilled in the art without creative work are all within the scope of protection of the present invention: Example 1 like Figure 1 As shown in the figure, the artificial intelligence-based swimming pool drowning prevention warning system includes: The acquisition module is used to send a basic questionnaire containing personal information, swimming experience, water activity preferences and physical characteristics to the user terminal, and estimate the user's breath holding time and training frequency based on the results of the basic questionnaire; the acquisition module includes multiple cameras, which are used to collect video data from any position on the water surface (such as Figure 2 shown).

[0024] Recognition module, including preset recognition models (such as Figure 2 The recognition model is used to obtain video data and obtain head features and water surface features in the video data through the recognition model, and obtain the entry and exit time of the head based on the relative position relationship between the facial features in the head features and the water surface features; the recognition module is also used to match personal information based on the head features to obtain the corresponding breath-holding time and training frequency range.

[0025] Among them, the user terminal is equipped with a SaaS membership management system. The administrator sends the basic questionnaire to the user terminal through the SaaS membership management system. The swimmer enters the filling results of the basic questionnaire through the SaaS membership management system of the user terminal and confirms the submission.

[0026] Among them, Figure 2 As shown, the recognition module is used to count according to the breath-holding time when the entry time is obtained. If the exit time is not obtained at the end of the timing, the video data and life-saving prompts are sent to the lifeguard terminal.

[0027] Among them, the identification module is also used to count the number of times the person enters the water during the breath-holding time according to personal information as the number of water entries, and obtain the water entry frequency based on the number of water entries. If the water entry frequency exceeds the training frequency range, video data and life-saving prompts are sent to the lifeguard terminal.

[0028] Among them, the recognition module is also used to obtain the interval time between two adjacent water entry times when the water entry frequency is within the training frequency range. If the interval time difference is greater than a preset difference (generally 3 seconds, which can be set by the administrator. The larger the interval time difference, the greater the change in the swimmer's water entry cycle, and the greater the risk of drowning; the specific interval time can also be set based on the swimmer's own swimming experience), video data and life-saving prompts are sent to the lifeguard terminal.

[0029] Among them, the recognition model includes a target detection model established by Faster R-CNN, SSD or YOLO. The recognition module uses a face dataset containing the positions of key facial features and a dataset containing various head postures to train the target detection model; the recognition model also includes a water surface segmentation model established by U-Net or DeepLab. The recognition model uses the water surface segmentation model to identify and segment the water surface in the video data, and then uses the gray-level co-occurrence matrix to extract the texture features of the water surface, and finally obtains the water surface features.

[0030] Among them, the recognition model is also used to identify the swimmer's behavioral information. The recognition module calculates optical flow using the Lucas-Kanade optical flow algorithm or the deep learning-based FlowNet method, analyzes the motion trajectory of the human head based on the calculated optical flow information, and obtains the swimmer's exit and entry behaviors based on the motion trajectory of the human head. The time corresponding to the exit and entry behaviors is recorded as the exit time and entry time, respectively.

[0031] The recognition module also includes a human posture estimation model built by OpenPose. The human posture estimation model infers the human posture by identifying the positions of key points such as the neck, shoulders, and torso nodes. The recognition model combines continuous posture estimation results into action sequences and uses recurrent neural networks or long short-term memory network models for action recognition. By analyzing the action sequence, the swimmer's continuous posture is judged as action behavior. Action behaviors include swimming, playing, and resting. Corresponding breath-holding time and training frequency ranges are set for the swimmer's different action behaviors.

[0032] Among them, when the recognition module matches personal information based on head features to obtain the corresponding breath-holding time and training frequency range, it first uses the target detection model to identify the swimmer's facial features, and then matches the swimmer's personal information based on the facial features; at the same time, it uses the human posture estimation model to infer the swimmer's body posture, and then obtains the swimmer's movement behavior based on the swimmer's continuous body posture; the swimmer's movement behavior and personal information are combined to obtain the swimmer's current breath-holding time and training frequency range.

[0033] Among them, the recognition module is used to obtain the number of other swimmers within the preset range of the swimmer (generally 1m, which can be set by the administrator or according to the swimmer's personal situation) when recognizing the play action. If the number is less than 1, the swimmer will be regarded as a dangerous swimmer, and the dangerous swimmer's action behavior, interval time, breath holding time and water entry frequency will be continuously monitored until other swimmers appear within the preset range of the dangerous swimmer; the recognition module is used to send the video data corresponding to the dangerous swimmer to the lifeguard terminal when a dangerous swimmer is identified.

[0034] Among them, the lifeguard terminal includes a prompt playback terminal and a video display terminal (multiple lifeguards can use one video display terminal, or each lifeguard can be equipped with a separate video display terminal, and the lifeguard's separate video display terminal is communicated with the prompt playback terminal). The video display terminal is used to play the received video data. The video display terminal stores a venue simulation map. The video display terminal is also used to obtain the collection position of the video data, and mark a warning red dot at the position corresponding to the collection position in the venue simulation map. The video display terminal is used to display the venue simulation map; the prompt playback terminal is used to play a life-saving prompt (generally a vibration prompt) when a life-saving signal is received, and highlight the warning red dot corresponding to the life-saving signal in the venue simulation map displayed on the video display terminal.

[0035] Among them, the lifeguard terminal includes several prompt playback terminals, each of which stores a lifeguard number and is provided with a confirmation button. The prompt playback terminal is used to use the information entered by the confirmation button as lifesaving confirmation information when playing the lifesaving prompt, and send the lifesaving confirmation information, the lifeguard number and the lifesaving prompt to the video display terminal. After receiving the confirmation information, the video display terminal stops flashing the warning red dot corresponding to the lifesaving prompt and counts down to a preset first arrival time (generally 30 seconds, which can be set by the administrator). If the lifeguard corresponding to the lifeguard number is captured in the video data corresponding to the lifesaving prompt before the countdown ends, the playback of the important lifesaving prompt and the video data corresponding to the important lifesaving prompt is stopped, the warning red dot corresponding to the important lifesaving prompt is deleted from the venue simulation map, and the countdown is stopped. If the lifeguard corresponding to the lifeguard number sent at the same time as the lifesaving prompt is not captured in the video data corresponding to the lifesaving prompt after the countdown ends, the important lifesaving prompt (a vibration prompt that is different from the lifesaving prompt) is sent to the prompt playback terminal according to the lifesaving prompt. After the video display terminal sends the important life-saving prompt, it flashes the warning red dot corresponding to the important life-saving prompt at a preset frequency (set by the administrator, generally with a visible difference from the flashing frequency of the life-saving signal, and generally faster than the flashing frequency of the life-saving signal); after receiving the important life-saving prompt, the prompt playback terminal plays the important life-saving prompt; until the appearance features of the lifeguard are collected in the video data corresponding to the important life-saving prompt, the playback of the important life-saving prompt and the video data corresponding to the important life-saving prompt is stopped, and the warning red dot corresponding to the important life-saving prompt is deleted from the venue simulation map.

[0036] Among them, the recognition module is also used to extract the swimmer's action behavior and pool position from the historically collected video data based on the swimmer's personal information, and use the extracted action behavior and pool position to establish a probability model; the recognition module is used to obtain the swimmer's current action behavior and current pool position from the currently received video data, and put the current action behavior and current pool position into the probability model to infer the rationality of the swimmer's current behavior. When the rationality is lower than the preset reasonable value, the swimmer is regarded as a dangerous swimmer.

[0037] This embodiment also provides an artificial intelligence-based swimming pool drowning prevention warning method using an artificial intelligence-based swimming pool drowning prevention warning system.

[0038] When implementing There are two lifeguards in the swimming pool of a swimming hall. Each lifeguard wears a reminder playback terminal (bracelet) with his or her personal number. There is a 2-square-meter video display terminal near the pool. There are 6 children and 3 adults accompanying the children in the children's area of ​​the swimming pool, 3 young people in training in the youth area, and 2 adults resting, 3 adults in training, and 3 adults playing in the adult area.

[0039] During the operation of the swimming pool, multiple cameras in the pool continuously collect water surface video data and transmit it to the recognition module.

[0040] In the children's area, Xiao Ming is an 8-year-old child, accompanied by his parents. Xiao Ming is playing in the shallow end of the pool. The system uses the object detection model to identify Xiao Ming's facial features and matches them with the basic questionnaire data he previously completed. It determines that he is a swimming beginner and is currently playing. Based on his personal information and movements, the system sets a corresponding breath-holding time and training frequency range for him. At this time, there are other children and adults within the preset range (1 meter) of Xiao Ming, so the system determines that he is in a normal state.

[0041] In the youth area, Xiao Li, a regular swimmer, was training for freestyle swimming. The recognition module used an optical flow algorithm and a human pose estimation model to accurately identify his entry and exit behavior, recording the entry and exit times and counting his frequency. Because his frequency was within his training range and the time interval between consecutive entries was stable, the system determined his behavior was normal.

[0042] In the adult area, Ms. Zhang was resting and Ms. Wang was doing routine swimming training. Their behavior was consistent with their respective patterns and frequency ranges, and did not trigger the system alarm. However, Mr. Chen was playing in the deep water, with no one else within 1 meter of him. After the recognition module detected this situation, it marked Mr. Chen as a dangerous swimmer and began continuously monitoring his movements, intervals, breath holding time, and frequency of water entry, while also sending his video data to the lifeguard terminal.

[0043] After receiving Mr. Chen's video data, the video display terminal at the lifeguard terminal begins playback. It also marks Mr. Chen's location (a red dot) on a simulated map of the venue based on the location where the video data was collected. The video display terminal then displays the simulated map and highlights the red dot, prompting the playback terminal (the wristbands worn by the two lifeguards) to vibrate upon receiving the rescue signal. The video display terminal also zooms in on Mr. Chen's video. Lifeguard Xiao Zhang first presses the confirmation button on the wristband to enter his rescue confirmation information. He then sends his number, rescue confirmation information, and a rescue prompt to the video display terminal. Upon receiving the confirmation, the video display terminal stops flashing the red dot corresponding to Mr. Chen's location (only highlighting it) and begins counting down to the preset first arrival time (30 seconds).

[0044] During these 30 seconds, Xiao Zhang quickly swam towards Mr. Chen's location. However, if Xiao Zhang fails to reach Mr. Chen's location within 30 seconds, and the video display terminal fails to capture Xiao Zhang's appearance, it will send an important life-saving reminder to the prompt playback terminal, which will vibrate again. The video display terminal will highlight and rapidly flash the warning red dot corresponding to Mr. Chen's location on the venue simulation map, prompting the lifeguard to speed up the rescue.

[0045] The recognition module also analyzed historical video data and the swimmers' personal information to determine the distribution probability of resting, swimming, and playing activities in adult areas for swimmers like Mr. Chen, as well as the probability of activity in different pool locations. This model then established a probability model. Mr. Chen's prolonged play in the deep water, unattended, was calculated within the probability model to be less plausible than the preset reasonable value, further confirming his dangerous state. This provided data support for the system's subsequent optimization of the drowning judgment criteria.

[0046] Xiao Zhang eventually arrived at Mr. Chen's side and confirmed he wasn't in danger of drowning. However, he provided safety education regarding the potential risks of playing alone in deep water, and Mr. Chen returned to the shallow end to continue his activities. This entire process demonstrated the operational mechanism and effectiveness of the AI-based swimming pool drowning prevention warning system in real-world scenarios, effectively ensuring the safety of those in the pool. Furthermore, through the analysis and utilization of historical data, the system's ability to assess drowning risks is continuously improved.

[0047] The above is only an embodiment of the present invention, and the common knowledge such as the specific structure and characteristics of the scheme is not described in detail here. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. The artificial intelligence-based swimming pool drowning prevention warning system is characterized by: include: A collection module is used to send a basic questionnaire data including personal information, swimming experience, water activity preferences and physical characteristics to the user terminal, and estimate the user's breath-holding time and training frequency based on the results of the basic questionnaire; the collection module includes multiple cameras, and the multiple cameras are used to collect video data at any position on the water surface; The recognition module includes a preset recognition model, the recognition model is used to obtain video data and obtain head features and water surface features in the video data through the recognition model, and obtain the time when the head enters the water and the time when the head exits the water according to the relative position relationship between the facial features in the head features and the water surface features; the recognition module is also used to match the personal information according to the head features to obtain the corresponding breath-holding time and training frequency range; The recognition module is used to count according to the breath-holding time when the water entry time is obtained, and if the water exit time is not obtained at the end of the timing, the video data and life-saving prompts are sent to the lifeguard terminal; The identification module is also used to count the number of times of entering the water during the breath-holding time according to the personal information as the number of times of entering the water, obtain the frequency of entering the water according to the number of times of entering the water, and send video data and lifesaving tips to the lifeguard terminal if the frequency of entering the water exceeds the training frequency range; The identification module is also used to obtain the interval time between two adjacent water immersion times when the water immersion frequency is within the training frequency range, and if the interval time difference is greater than a preset difference, send video data and lifesaving tips to the lifeguard terminal.

2. The artificial intelligence-based swimming pool drowning prevention warning system according to claim 1, characterized in that: The recognition model includes a target detection model established by Faster R-CNN, SSD or YOLO, and the recognition module uses a face data set containing the positions of key points of facial features and a data set containing various postures of the head to train the target detection model; the recognition model also includes a water surface segmentation model established by U-Net or DeepLab, and the recognition model uses the water surface segmentation model to identify and segment the water surface in the video data, and then uses the gray-level co-occurrence matrix to extract the texture features of the water surface, and finally obtains the water surface features.

3. The artificial intelligence-based swimming pool drowning prevention warning system according to claim 2, characterized in that: The recognition model is also used to recognize the behavior information of the swimmer. The recognition module calculates the optical flow by using the Lucas-Kanade optical flow algorithm or the FlowNet method based on deep learning, analyzes the motion trajectory of the head according to the calculated optical flow information, obtains the swimmer's water exit and water entry behaviors according to the motion trajectory of the head, and records the times corresponding to the water exit and water entry behaviors as the water exit time and water entry time, respectively. The recognition module also includes a human posture estimation model constructed by OpenPose, which infers the posture of the human body by identifying the positions of the key points of the human body, namely the neck, shoulders, and torso nodes. The recognition model combines continuous posture estimation results into an action sequence, and uses a recurrent neural network or a long short-term memory network model to perform action recognition. By analyzing the action sequence, the swimmer's continuous posture is judged as an action behavior, and the action behavior includes swimming action, playing action, and resting action, and corresponding breath holding time and training frequency range are set for different action behaviors of the swimmer.

4. The artificial intelligence-based swimming pool drowning prevention warning system according to claim 3 is characterized by: When the recognition module matches the personal information according to the head features to obtain the corresponding breath-holding time and training frequency range, the target detection model is first used to recognize the facial features of the swimmer, and the personal information of the swimmer is matched according to the facial features; at the same time, the human body posture estimation model is used to infer the swimmer's body posture, and then the swimmer's movement behavior is obtained according to the swimmer's continuous body posture; the swimmer's movement behavior and personal information are combined to obtain the swimmer's current breath-holding time and training frequency range.

5. The artificial intelligence-based swimming pool drowning prevention warning system according to claim 4 is characterized in that: The recognition module is used to obtain the number of other swimmers within the preset range of the swimmer when the play action is recognized. If the number is less than 1, the swimmer is regarded as a dangerous swimmer, and the action behavior, interval time, breath holding time and water entry frequency of the dangerous swimmer are continuously monitored until other swimmers appear within the preset range of the dangerous swimmer; The identification module is used to send the video data corresponding to the dangerous swimmer to the lifeguard terminal when a dangerous swimmer is identified.

6. The artificial intelligence-based swimming pool drowning prevention warning system according to claim 5 is characterized by: The lifeguard terminal includes a prompt playback terminal and a video display terminal. The video display terminal is used to play the received video data. The video display terminal stores a venue simulation map. The video display terminal is also used to obtain the collection position of the video data and mark a warning red dot at a position corresponding to the collection position in the venue simulation map. The video display terminal is used to display the venue simulation map. The prompt playback terminal is used to play a life-saving prompt when a life-saving signal is received, and highlight and flash the warning red dot corresponding to the life-saving signal in the venue simulation map displayed on the video display terminal.

7. The artificial intelligence-based swimming pool drowning prevention warning system according to claim 6 is characterized by: The lifeguard terminal comprises a plurality of prompt playback terminals, wherein the prompt playback terminal stores a lifeguard number, and a confirmation button is provided on the prompt playback terminal; the prompt playback terminal is used to use the information entered by the confirmation button as the lifesaving confirmation information when playing the lifesaving prompt, and send the lifesaving confirmation information, the lifeguard number and the lifesaving prompt to the video display terminal, and the video display terminal stops flashing the warning red dot corresponding to the lifesaving prompt after receiving the confirmation information, and counts down the preset first arrival time; if the lifeguard corresponding to the lifeguard number is collected in the video data corresponding to the lifesaving prompt before the countdown ends, then the important lifesaving prompt and the video data corresponding to the important lifesaving prompt are stopped from playing, the warning red dot corresponding to the important lifesaving prompt is deleted from the venue simulation diagram, and the countdown is stopped; if after the countdown to the first arrival time ends, the lifeguard corresponding to the lifeguard number sent at the same time as the lifesaving prompt is not collected in the video data corresponding to the lifesaving prompt, then the important lifesaving prompt is sent to the prompt playback terminal according to the lifesaving prompt; After the video display terminal sends the important life-saving prompt, the warning red dot corresponding to the important life-saving prompt flashes at a preset frequency; after the prompt playback terminal receives the important life-saving prompt, it plays the important life-saving prompt; Until the appearance features of the lifeguard are collected in the video data corresponding to the important life-saving reminder, stop playing the important life-saving reminder and the video data corresponding to the important life-saving reminder, and delete the warning red dot corresponding to the important life-saving reminder from the venue simulation map.

8. The artificial intelligence-based swimming pool drowning prevention warning system according to claim 7, characterized in that: The recognition module is also used to extract the swimmer's action behavior and swimming pool position from historically collected video data based on the swimmer's personal information, and use the extracted action behavior and swimming pool position to establish a probability model; the recognition module is used to obtain the swimmer's current action behavior and current swimming pool position from the currently received video data, and put the current action behavior and current swimming pool position into the probability model to infer the rationality of the swimmer's current behavior, and when the rationality is lower than a preset rational value, the swimmer is regarded as a dangerous swimmer.

9. A swimming pool drowning prevention warning method based on artificial intelligence, characterized in that: The artificial intelligence-based swimming pool drowning prevention warning system described in any one of claims 1 to 8 is used.

Images