Swimming pool drowning video monitoring system based on lidar

Abstract

The present invention relates to the field of swimming pool drowning video monitoring, specifically a laser radar-based swimming pool drowning video monitoring system, which includes data acquisition, laser radar, video camera, alarm output module and database. The data acquisition module builds a swimmer platform and collects data. The laser radar module uses the pool wall as a reference to obtain target point data, construct a three-dimensional image, calculate the head distance difference, the slope of the tangent line between the limbs and the abdomen, and the posture abnormality index, determine and send a first warning signal; after receiving the signal, the video camera module analyzes the human body posture and trajectory changes, uses the convolutional neural network drowning model to predict the drowning probability, and generates a second warning signal; after receiving the second warning signal, the alarm output module locates the drowning person, plans a rescue route, and notifies the large-screen display and lifeguard wristband; the present invention integrates multiple modules to achieve accurate monitoring and rapid response to swimming pool drowning situations, effectively ensuring swimming pool safety.

Description

Technical Field

[0001] The present invention relates to the field of swimming pool drowning video monitoring, and in particular to a swimming pool drowning video monitoring system based on laser radar. Background Art

[0002] In the development of swimming pool drowning detection technology, traditional monitoring methods have gradually exposed serious limitations. Pure video monitoring technology has achieved automated monitoring to a certain extent, but it faces many difficult-to-overcome challenges. For example, the refraction, reflection, and fluctuation of water bodies can interfere with the quality of video images, blurring the details of swimmers' movements and increasing the difficulty of accurately identifying drowning behavior, significantly reducing the accuracy of pure video monitoring systems.

[0003] The emergence of LiDAR technology has brought new opportunities for swimming pool drowning monitoring. In a swimming pool environment, LiDAR can continuously scan targets, acquiring a large amount of target point data including distance and reflection intensity, and then constructing a three-dimensional image of the pool. This provides depth information and spatial position data for precise monitoring, effectively compensating for the depth perception shortcomings of video monitoring. RGB cameras play an indispensable role in visual information collection in monitoring systems. They can capture high-definition images of the pool and record the movement trajectories of swimmers, providing intuitive data support for the system from a visual perspective.

[0004] However, practical applications of either LiDAR or video camera technology alone still face challenges. While LiDAR offers advantages in distance and speed measurement, it is relatively weak in recognizing complex human posture details and subtle changes in movement. Furthermore, the accuracy of image analysis by video cameras needs to be improved when faced with water interference and complex backgrounds.

[0005] The deep integration of lidar and video camera groups to build a new lidar-based swimming pool drowning video monitoring system has extremely important practical significance and application value. Summary of the Invention

[0006] In order to solve the technical problems raised by the above background technology, the present invention provides a swimming pool drowning video monitoring system based on laser radar.

[0007] The purpose of the present invention can be achieved through the following technical solutions:

[0008] The present invention is a swimming pool drowning video monitoring system based on laser radar, which includes a data acquisition module, a laser radar module, a video camera module, an alarm output module and a database.

[0009] The data acquisition module is used to build a swimmer platform and install a lidar sensor and RGB camera group in the swimming pool. It collects basic information data of people in the swimming pool and motion data in real time, and sends them to the lidar module and video camera module respectively. The specific process is as follows:

[0010] The data acquisition module is provided with a platform establishment unit and an acquisition unit;

[0011] The platform establishment unit is used to build a swimmer's data platform, through which swimmers upload their personal information. Specifically, the platform establishes a swimmer login interface through the display screen. The swimmer login interface includes inputting weight, number of swimming certificates, number of monthly swimming times, single swimming distance, and single swimming time. When the swimmer completes the input on the login interface, a user interface corresponding to the swimmer is established and a new identity number for the swimmer is generated and sent to the database for storage.

[0012] The basic data information uploaded by each swimmer is parsed in turn, and the selected data of weight, number of swimming certificates, number of monthly swimming times, single swimming distance and single swimming time are marked as stw1, stw2, stw3, stw4 and stw5 respectively, and then normalized and entered into the formula for calculation. Obtain a comprehensive swimming ability assessment value for each swimmer, where Q1, Q2, Q3, Q4, and Q5 are weight coefficients corresponding to weight, number of swimming certificates, number of monthly swimming times, single swimming distance, and single swimming time; i represents the swimmer's number, and n represents the number of swimmers. Synchronously send the corresponding comprehensive swimming ability assessment value for each swimmer to a database for storage;

[0013] The acquisition unit divides the deep pool and the shallow pool into several swimming areas in equal proportions. RGB cameras are installed according to the number of swimming areas. One RGB camera collects data information of each swimmer in one swimming area and sends the acquired data information to the video camera module. A lidar sensor is installed directly above the center of the swimming pool to collect target point data of each swimmer in real time and send it to the lidar module.

[0014] The LiDAR module emits near-infrared light signals and measures target point data to construct a three-dimensional image of the pool. It then identifies the swimmer's position in the pool, calculates the difference, slope, and posture abnormality index, and determines the swimmer's first warning signal. The specific process is as follows:

[0015] The pool wall is used as a reference point. Near-infrared light signals are emitted in real time by a lidar (LiDAR) to contact various targets, including swimmers, floatation equipment, and the pool wall. This data is used to obtain a large amount of target point data in the pool, including the distance and reflection intensity of the target. This target point data is then filtered to remove noise and outliers. The lidar position is used as the origin of the reference coordinate system, and the processed target point data is used to construct a three-dimensional image of the pool.

[0016] Real-time identification of the head target point positions, limb target point positions, and abdomen target point positions of each swimmer in the three-dimensional image of the swimming pool. The head target point positions are connected to the water surface positions to obtain the actual head distance value between each swimmer's head and the water surface. The standard head distance value in the database is extracted, and the actual head distance value is subtracted from the standard head distance value to obtain the head distance difference value.

[0017] A two-dimensional rectangular coordinate system is constructed with the acquisition time point as the horizontal coordinate and the position of each limb target point and the position of each abdominal target point as the vertical coordinate, and the position of each limb target point and the position of each abdominal target point are sequentially input into the coordinate system in the order of the acquisition time point, and the position of each limb target point and the position of each abdominal target point are marked as limb movement points and abdominal movement points in the coordinate system, and smooth curves are used to connect the limb movement points and the abdominal movement points in sequence to obtain a curve diagram of the limb movement points and the abdominal movement points changing with time, and tangents of the curves are drawn at the limb movement points and the abdominal movement points to obtain the limb movement tangent and the abdominal movement tangent, and the slopes of the limb movement tangent and the abdominal movement tangent are calculated to obtain the limb movement tangent slope and the abdominal movement tangent slope;

[0018] The head distance difference, limb movement tangent slope and abdomen movement tangent slope are marked as GY, GR and GE respectively, and after normalization, they are entered into the formula for calculation The swimmer's posture abnormality index GH, where J1, J2 and J3 are weight coefficients corresponding to the head distance difference, the tangent slope of the limb movement and the tangent slope of the abdomen movement; extract the preset posture abnormality range in the database, if the posture abnormality index is less than the minimum value of the posture abnormality range, the swimmer is marked as a non-dangerous person, if the posture abnormality index is within the posture abnormality range, the swimmer is marked as a person to be analyzed, if the posture abnormality index is greater than the maximum value within the posture abnormality range, the swimmer is marked as a dangerous person, and a first warning signal is generated and sent to the video camera module, and the first warning signal includes the location information of the dangerous person and the person to be analyzed.

[0019] The video camera module analyzes the swimmer's posture and trajectory changes in the video captured by the RGB camera group. It determines whether the swimmer's movement is abnormal based on the preset drowning model and generates a second warning signal accordingly. The specific process is as follows:

[0020] The video camera module receives the first warning signal and obtains the position information of the dangerous person and the person to be analyzed through the RGB camera group. It marks the dangerous person as an orange frame in the real-time video and the person to be analyzed as a yellow frame. It then identifies the real-time status of the dangerous person and the person to be analyzed in turn. Specifically, it first obtains the video image of the dangerous person in the swimming pool, decomposes the video image frame by frame into continuous video frames, and marks any adjacent position points of the dangerous person in the continuous video frames as P1 (x1, y1) and P2 (x2, y2). Then, it obtains the time interval of the adjacent position points as t1, and calculates the time interval using the formula Get the trajectory speed CW of the dangerous person;

[0021] Set the parametric equations of the trajectory to f = f(t2) and g = g(t2), and use the formula to calculate The curvature CR of the dangerous person's trajectory is obtained, where f 1 (t2), f 2 (t2), g 1 (t2) and g 2 (t2) is the first and second derivatives of f = f(t2) and g = g(t2);

[0022] Identify the coordinates of adjacent joints of the dangerous person in sequence, connect the coordinates of three adjacent joints to obtain a joint triangle, including a joint triangle formed by connecting the shoulder, elbow, and wrist joints. With the joint triangle as the center, obtain the vertical displacement values ​​of the joint triangles in consecutive video frames, and calculate the average of the displacement values ​​to obtain the average displacement value.

[0023] A deep learning convolutional neural network is selected to build a drowning model, specifically:

[0024] A convolutional neural network is used as the architecture of the drowning model, which includes a convolutional layer, a pooling layer, and a fully connected layer. The sum of the feature quantities of the dangerous person's trajectory speed, trajectory curvature, comprehensive evaluation value of swimming ability, and average displacement value is used as the number of convolutional layer nodes. Drowning case data and normal swimming behavior data in the database are extracted and integrated into a drowning dataset. The drowning model is trained with the drowning dataset, which is divided into a training set and a test set. 80% of the drowning dataset is used as the training set and 20% as the test set. The trajectory speed, trajectory curvature, comprehensive evaluation value of swimming ability, and average displacement value of any dangerous person are used as input feature quantities of the convolutional layer, and the drowning probability prediction value is used as the output feature value of the fully connected layer. The back-propagation algorithm is used to calculate the drowning model. And iterate in sequence until the drowning probability prediction value is obtained; extract the preset drowning probability prediction range in the database, if the drowning probability prediction value is greater than the maximum value of the preset drowning probability prediction range, the dangerous person is marked as a red frame, and a rescue instruction is generated and sent to the alarm output module; if the drowning probability prediction value is less than the minimum value of the preset drowning probability prediction range, the dangerous person is marked as a blue frame, and the dangerous person mark is removed; and so on, the trajectory speed, trajectory curvature and average displacement value of the person to be analyzed are input into the drowning model to obtain the drowning probability prediction value of the person to be analyzed. If it exceeds the preset threshold, a second warning signal is generated until all dangerous persons and persons to be analyzed are marked as blue frames. It should be noted that being marked as a blue frame indicates that the swimmer's movement state is normal.

[0025] The alarm output module receives the second warning signal, obtains the location information of the dangerous person and the video image of the drowning person, sends them to the large-screen display, and vibrates the lifeguard's smart bracelet. The specific process is as follows:

[0026] Upon receiving the second warning signal, the drowning time of the drowning person is measured, and the position of the drowning person is used as the center point to obtain the position information of the lifeguards within the radius. The position of the drowning person and the lifeguards are connected to obtain the rescue route, and the rescue routes are sorted from large to small. The lifeguard corresponding to the smallest rescue route is marked as the preferred person, and the video image and position information of the drowning person are sent to the large-screen display screen. At the same time, the smart bracelet of the preferred person is vibrated; if the preferred person has been assigned a rescue task, the smart bracelet of the second-ranked lifeguard is vibrated until the rescue is completed.

[0027] Compared with the existing technology, the beneficial effects of the present invention are: precise drowning monitoring capability: by emitting near-infrared light signals and measuring target point data through the laser radar module, a three-dimensional image of the swimming pool is constructed, which can accurately identify various parts of the swimmer in the swimming pool, calculate key indicators such as the head distance difference, the tangent slopes of the limbs and abdomen, and the posture abnormality index, so as to accurately determine the swimmer's dangerous state and issue the first warning signal in time; the video camera module uses a deep learning convolutional neural network drowning model, combined with multi-dimensional data such as the swimmer's trajectory speed, trajectory curvature, comprehensive evaluation value of swimming ability and average displacement value for analysis, further improving the accuracy of drowning situation judgment and effectively reducing the risk of false alarms and missed alarms.

[0028] Efficient rescue response mechanism: After receiving the second warning signal, the alarm output module can quickly obtain the location information and video images of the drowning person and send them to the large-screen display, so that managers and relevant staff can intuitively understand the on-site situation; through intelligent algorithms, the positions of lifeguards within the radius are quickly determined, rescue routes are planned, and preferred personnel are screened, and their smart bracelets are vibrated to remind them to ensure that the rescue operation can be started quickly; if the preferred person already has a task, it can also automatically switch to the lifeguard at the bottom of the list, ensuring the timeliness and efficiency of the rescue and increasing the chance of rescuing the drowning person. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. The following drawings are not intentionally scaled to the actual size, and the focus is on illustrating the main purpose of the present invention.

[0030] Figure 1 This is a principle block diagram of the present invention. DETAILED DESCRIPTION

[0031] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts also fall within the scope of protection of the present invention.

[0032] Please refer to Figure 1 As shown, the present invention is a swimming pool drowning video monitoring system based on laser radar, which includes a data acquisition module, a laser radar module, a video camera module, an alarm output module and a database.

[0033] The data acquisition module is used to build a swimmer platform and install a lidar sensor and RGB camera group in the swimming pool. It collects basic information data of people in the swimming pool and motion data in real time, and sends them to the lidar module and video camera module respectively. The specific process is as follows:

[0034] The data acquisition module is provided with a platform establishment unit and an acquisition unit;

[0035] The platform establishment unit is used to build a swimmer's data platform, through which swimmers upload their personal information. Specifically, the platform establishes a swimmer login interface through the display screen. The swimmer login interface includes inputting weight, number of swimming certificates, number of monthly swimming times, single swimming distance, and single swimming time. When the swimmer completes the input on the login interface, a user interface corresponding to the swimmer is established and a new identity number for the swimmer is generated and sent to the database for storage.

[0036] The basic data information uploaded by each swimmer is parsed in turn, and the selected data of weight, number of swimming certificates, number of monthly swimming times, single swimming distance and single swimming time are marked as stw1, stw2, stw3, stw4 and stw5 respectively, and then normalized and entered into the formula for calculation. Obtain a comprehensive swimming ability assessment value for each swimmer, where Q1, Q2, Q3, Q4, and Q5 are weight coefficients corresponding to weight, number of swimming certificates, number of monthly swimming times, single swimming distance, and single swimming time; i represents the swimmer's number, and n represents the number of swimmers. Synchronously send the corresponding comprehensive swimming ability assessment value for each swimmer to a database for storage;

[0037] The acquisition unit divides the deep pool and the shallow pool into several swimming areas in equal proportions. RGB cameras are installed according to the number of swimming areas. One RGB camera collects data information of each swimmer in one swimming area and sends the acquired data information to the video camera module. A lidar sensor is installed directly above the center of the swimming pool to collect target point data of each swimmer in real time and send it to the lidar module.

[0038] The LiDAR module emits near-infrared light signals and measures target point data to construct a three-dimensional image of the pool. It then identifies the swimmer's position in the pool, calculates the difference, slope, and posture abnormality index, and determines the swimmer's first warning signal. The specific process is as follows:

[0039] The pool wall is used as a reference point. Near-infrared light signals are emitted in real time by a lidar (LiDAR) to contact various targets, including swimmers, floatation equipment, and the pool wall. This data is used to obtain a large amount of target point data in the pool, including the distance and reflection intensity of the target. This target point data is then filtered to remove noise and outliers. The lidar position is used as the origin of the reference coordinate system, and the processed target point data is used to construct a three-dimensional image of the pool.

[0040] Real-time identification of the head target point positions, limb target point positions, and abdomen target point positions of each swimmer in the three-dimensional image of the swimming pool. The head target point positions are connected to the water surface positions to obtain the actual head distance value between each swimmer's head and the water surface. The standard head distance value in the database is extracted, and the actual head distance value is subtracted from the standard head distance value to obtain the head distance difference value.

[0041] A two-dimensional rectangular coordinate system is constructed with the acquisition time point as the horizontal coordinate and the position of each limb target point and the position of each abdominal target point as the vertical coordinate, and the position of each limb target point and the position of each abdominal target point are sequentially input into the coordinate system in the order of the acquisition time point, and the position of each limb target point and the position of each abdominal target point are marked as limb movement points and abdominal movement points in the coordinate system, and smooth curves are used to connect the limb movement points and the abdominal movement points in sequence to obtain a curve diagram of the limb movement points and the abdominal movement points changing with time, and tangents of the curves are drawn at the limb movement points and the abdominal movement points to obtain the limb movement tangent and the abdominal movement tangent, and the slopes of the limb movement tangent and the abdominal movement tangent are calculated to obtain the limb movement tangent slope and the abdominal movement tangent slope;

[0042] The head distance difference, limb movement tangent slope and abdomen movement tangent slope are marked as GY, GR and GE respectively, and after normalization, they are entered into the formula for calculation The swimmer's posture abnormality index GH, where J1, J2 and J3 are weight coefficients corresponding to the head distance difference, the tangent slope of the limb movement and the tangent slope of the abdomen movement; extract the preset posture abnormality range in the database, if the posture abnormality index is less than the minimum value of the posture abnormality range, the swimmer is marked as a non-dangerous person, if the posture abnormality index is within the posture abnormality range, the swimmer is marked as a person to be analyzed, if the posture abnormality index is greater than the maximum value within the posture abnormality range, the swimmer is marked as a dangerous person, and a first warning signal is generated and sent to the video camera module, and the first warning signal includes the location information of the dangerous person and the person to be analyzed.

[0043] The video camera module analyzes the swimmer's posture and trajectory changes in the video captured by the RGB camera group. It determines whether the swimmer's movement is abnormal based on the preset drowning model and generates a second warning signal accordingly. The specific process is as follows:

[0044] The video camera module receives the first warning signal and obtains the position information of the dangerous person and the person to be analyzed through the RGB camera group. It marks the dangerous person as an orange frame in the real-time video and the person to be analyzed as a yellow frame. It then identifies the real-time status of the dangerous person and the person to be analyzed in turn. Specifically, it first obtains the video image of the dangerous person in the swimming pool, decomposes the video image frame by frame into continuous video frames, and marks any adjacent position points of the dangerous person in the continuous video frames as P1 (x1, y1) and P2 (x2, y2). Then, it obtains the time interval of the adjacent position points as t1, and calculates the time interval using the formula Get the trajectory speed CW of the dangerous person;

[0045] Set the parametric equations of the trajectory to f = f(t2) and g = g(t2), and use the formula to calculate The curvature CR of the dangerous person's trajectory is obtained, where f 1 (t2), f 2 (t2), g 1 (t2) and g 2 (t2) is the first and second derivatives of f = f(t2) and g = g(t2);

[0046] Identify the coordinates of adjacent joints of the dangerous person in sequence, connect the coordinates of three adjacent joints to obtain a joint triangle, including a joint triangle formed by connecting the shoulder, elbow, and wrist joints. With the joint triangle as the center, obtain the vertical displacement values ​​of the joint triangles in consecutive video frames, and calculate the average of the displacement values ​​to obtain the average displacement value.

[0047] A deep learning convolutional neural network is selected to build a drowning model, specifically:

[0048] A convolutional neural network is used as the architecture of the drowning model, which includes a convolutional layer, a pooling layer, and a fully connected layer. The sum of the feature quantities of the dangerous person's trajectory speed, trajectory curvature, comprehensive evaluation value of swimming ability, and average displacement value is used as the number of convolutional layer nodes. Drowning case data and normal swimming behavior data in the database are extracted and integrated into a drowning dataset. The drowning model is trained with the drowning dataset, which is divided into a training set and a test set. 80% of the drowning dataset is used as the training set and 20% as the test set. The trajectory speed, trajectory curvature, comprehensive evaluation value of swimming ability, and average displacement value of any dangerous person are used as input feature quantities of the convolutional layer, and the drowning probability prediction value is used as the output feature value of the fully connected layer. The back-propagation algorithm is used to calculate the drowning model. And iterate in sequence until the drowning probability prediction value is obtained; extract the preset drowning probability prediction range in the database, if the drowning probability prediction value is greater than the maximum value of the preset drowning probability prediction range, the dangerous person is marked as a red frame, and a rescue instruction is generated and sent to the alarm output module; if the drowning probability prediction value is less than the minimum value of the preset drowning probability prediction range, the dangerous person is marked as a blue frame, and the dangerous person mark is removed; and so on, the trajectory speed, trajectory curvature and average displacement value of the person to be analyzed are input into the drowning model to obtain the drowning probability prediction value of the person to be analyzed. If it exceeds the preset threshold, a second warning signal is generated until all dangerous persons and persons to be analyzed are marked as blue frames. It should be noted that being marked as a blue frame indicates that the swimmer's movement state is normal.

[0049] The alarm output module receives the second warning signal, obtains the location information of the dangerous person and the video image of the drowning person, sends them to the large-screen display, and vibrates the lifeguard's smart bracelet. The specific process is as follows:

[0050] Upon receiving the second warning signal, the drowning time of the drowning person is measured, and the position of the drowning person is used as the center point to obtain the position information of the lifeguards within the radius. The position of the drowning person and the lifeguards are connected to obtain the rescue route, and the rescue routes are sorted from large to small. The lifeguard corresponding to the smallest rescue route is marked as the preferred person, and the video image and position information of the drowning person are sent to the large-screen display screen. At the same time, the smart bracelet of the preferred person is vibrated; if the preferred person has been assigned a rescue task, the smart bracelet of the second-ranked lifeguard is vibrated until the rescue is completed.

[0051] The above is an illustration of the present invention and should not be considered as limiting thereof. Although several exemplary embodiments of the present invention have been described, it will be readily understood by those skilled in the art that many modifications may be made to the exemplary embodiments without departing from the novel teachings and advantages of the present invention. Therefore, all such modifications are intended to be included within the scope of the present invention as defined by the claims. It should be understood that the above is an illustration of the present invention and should not be considered as being limited to the specific embodiments disclosed, and modifications to the disclosed embodiments and other embodiments are intended to be included within the scope of the appended claims. The present invention is defined by the claims and their equivalents.

Claims

1. A swimming pool drowning video monitoring system based on laser radar, including a data acquisition module, a video camera module, an alarm output module and a database, characterized in that: Also includes a lidar module; The video camera module analyzes the swimmer's posture and trajectory changes in the video captured by the RGB camera group, determines whether the swimmer's movement is abnormal based on the preset drowning model, and generates a second warning signal accordingly; The alarm output module receives the second warning signal, obtains the location information of the corresponding dangerous person and the video image of the drowning person, sends them to the large-screen display, and vibrates the lifeguard's smart bracelet; The LiDAR module emits near-infrared light signals and measures target point data to construct a three-dimensional image of the pool, identify the swimmer's position in the pool, and calculate the difference and slope. Specifically: The pool wall is used as a reference point. The laser radar emits near-infrared light signals in real time to contact various targets and return them to the laser radar. This generates a large amount of target point data in the pool. The target point data is filtered to remove noise and outliers. The laser radar position is used as the origin of the reference coordinate system, and the processed target point data is used to construct a three-dimensional image of the pool. The system identifies the target points of each swimmer's head, limbs, and abdomen in real time within the three-dimensional image of the swimming pool. The head target point positions are connected to the water surface positions to obtain the actual head distance between each swimmer's head and the water surface. The standard head distance value in the database is extracted and the actual head distance value is subtracted from the standard head distance value to obtain the head distance difference value. A two-dimensional rectangular coordinate system is constructed with the acquisition time point as the horizontal coordinate and the position of each limb target point and each abdominal target point as the vertical coordinate, and the position of each limb target point and each abdominal target point are sequentially input into the coordinate system in the order of the acquisition time point. The position of each limb target point and each abdominal target point are marked as limb moving points and abdominal moving points in the coordinate system, and smooth curves are used to connect the limb moving points and the abdominal moving points in sequence to obtain a curve diagram of the limb moving points and the abdominal moving points changing with time. Tangents of the curves are drawn at the limb moving points and the abdominal moving points to obtain the limb moving tangent and the abdominal moving tangent. At the same time, the slopes of the limb moving tangent and the abdominal moving tangent are calculated to obtain the slopes of the limb moving tangent and the abdominal moving tangent.

2. The laser radar-based swimming pool drowning video monitoring system according to claim 1 is characterized in that: The laser radar module calculates and obtains the abnormal posture index based on the obtained data, and determines the first warning signal of the swimmer, specifically: The head distance difference, limb movement tangent slope and abdomen movement tangent slope are marked as GY, GR and GE respectively, and after normalization, they are entered into the formula for calculation The swimmer's posture abnormality index GH, where J1, J2 and J3 are weight coefficients corresponding to the head distance difference, the tangent slope of the limb movement and the tangent slope of the abdomen movement; extract the preset posture abnormality range in the database, if the posture abnormality index is less than the minimum value of the posture abnormality range, the swimmer is marked as a non-dangerous person, if the posture abnormality index is within the posture abnormality range, the swimmer is marked as a person to be analyzed, if the posture abnormality index is greater than the maximum value within the posture abnormality range, the swimmer is marked as a dangerous person, and a first warning signal is generated and sent to the video camera module, and the first warning signal includes the location information of the dangerous person and the person to be analyzed.

3. The laser radar-based swimming pool drowning video monitoring system according to claim 1 is characterized in that: The video camera module analyzes the trajectory changes of the swimmer in the video based on the video captured by the RGB camera group to obtain the trajectory speed and trajectory curvature of the dangerous person. The specific process is as follows: After receiving the first warning signal, the position information of the dangerous person and the person to be analyzed is obtained through the RGB camera group, and the dangerous person is marked as an orange frame in the real-time video, and the person to be analyzed is marked as a yellow frame. The real-time status of the dangerous person and the person to be analyzed are identified in turn. Specifically, the video image of the dangerous person in the swimming pool is first obtained, and the video image is decomposed into continuous video frames frame by frame. Any adjacent position points of the dangerous person in the continuous video frame are marked as P1(x1, y1) and P2(x2, y2). Then, the time interval of the adjacent position points is obtained as t1, and the formula is used to calculate Get the trajectory speed CW of the dangerous person; Set the parametric equations of the trajectory to f = f(t2) and g = g(t2), and use the formula to calculate The curvature CR of the dangerous person's trajectory is obtained, where f 1 (t2), f 2 (t2), g 1 (t2) and g 2 (t2) is the first and second derivatives of f = f(t2) and g = g(t2).

4. The laser radar-based swimming pool drowning video monitoring system according to claim 3 is characterized in that: The video camera module analyzes the human body posture and obtains the average displacement value, specifically: Based on a preset drowning model, it is determined whether the swimmer's movement is abnormal, and a second warning signal is generated accordingly. The coordinate positions of adjacent joint points of the dangerous person are identified in sequence, and the coordinates of the three adjacent joint points are connected to obtain a joint triangle, including a joint triangle formed by connecting the shoulder, elbow and wrist joint points. With this joint triangle as the center, the displacement values ​​of the joint triangle in the vertical direction in continuous video frames are obtained, and the average displacement value is calculated by averaging the displacement values.

5. The laser radar-based swimming pool drowning video monitoring system according to claim 3 is characterized in that: The video camera module selects a deep learning convolutional neural network to construct a drowning model and obtains a drowning probability prediction value, specifically: A convolutional neural network is used as the architecture of the drowning model, which includes a convolutional layer, a pooling layer, and a fully connected layer. The sum of the feature numbers of the dangerous person's trajectory speed, trajectory curvature, comprehensive evaluation value of swimming ability, and average displacement value is used as the number of convolutional layer nodes. Drowning case data and normal swimming behavior data in the database are extracted and integrated into a drowning dataset. The drowning model is trained with the drowning dataset, which is divided into a training set and a test set. 80% of the drowning dataset is used as the training set and 20% as the test set. The trajectory speed, trajectory curvature, and average displacement of any dangerous person are used as the training set and the rest are used as the test set. , the comprehensive evaluation value of swimming ability and the average displacement value are used as the input feature values ​​of the convolution layer, and the drowning probability prediction value is used as the output feature value of the fully connected layer; the back propagation algorithm is used to calculate and iterate in sequence until the drowning probability prediction value is obtained; the preset drowning probability prediction range in the database is extracted, and if the drowning probability prediction value is greater than the maximum value of the preset drowning probability prediction range, the dangerous person is marked as a red frame, and a rescue instruction is generated and sent to the alarm output module; if the drowning probability prediction value is less than the minimum value of the preset drowning probability prediction range, the dangerous person is marked as a blue frame, and the dangerous person mark is removed; By analogy, the trajectory speed, trajectory curvature and average displacement value of the person to be analyzed are input into the drowning model to obtain the predicted value of the drowning probability of the person to be analyzed. If it exceeds the preset threshold, a second warning signal is generated until all dangerous persons and persons to be analyzed are marked as blue frames.

6. The laser radar-based swimming pool drowning video monitoring system according to claim 1 is characterized in that: The alarm output module receives the second warning signal, obtains the location information of the corresponding dangerous person and the video image of the drowning person, sends them to the large-screen display, and vibrates the lifeguard's smart bracelet. The specific process is as follows: Upon receiving the second warning signal, the drowning time of the drowning person is measured, and the position of the drowning person is used as the center point to obtain the position information of the lifeguards within the radius. The position of the drowning person and the lifeguards are connected to obtain the rescue route, and the rescue routes are sorted from large to small. The lifeguard corresponding to the smallest rescue route is marked as the preferred person, and the video image and position information of the drowning person are sent to the large-screen display screen. At the same time, the smart bracelet of the preferred person is vibrated; if the preferred person has been assigned a rescue task, the smart bracelet of the second-ranked lifeguard is vibrated until the rescue is completed.

7. The laser radar-based swimming pool drowning video monitoring system according to claim 1, characterized in that: The data acquisition module is used to establish a swimmer platform and obtain a comprehensive evaluation value of each swimmer's swimming ability. The specific process is as follows: The platform establishment unit is used to build a swimmer's data platform, through which swimmers upload their personal information. Specifically, the platform establishes a swimmer login interface through the display screen. The swimmer login interface includes inputting weight, number of swimming certificates, number of monthly swimming times, single swimming distance, and single swimming time. When the swimmer completes the input on the login interface, a user interface corresponding to the swimmer is established and a new identity number for the swimmer is generated and sent to the database for storage. The basic data information uploaded by each swimmer is parsed in turn, and the selected data of weight, number of swimming certificates, number of monthly swimming times, single swimming distance and single swimming time are marked as stw1, stw2, stw3, stw4 and stw5 respectively, and then normalized and entered into the formula for calculation. A comprehensive evaluation value of the swimming ability of each swimmer is obtained, where Q1, Q2, Q3, Q4, and Q5 are weight coefficients corresponding to weight, number of swimming certificates, number of monthly swimming times, single swimming distance, and single swimming time. i represents the number of each swimmer, and n represents the number of swimmers. The comprehensive evaluation value of the swimming ability corresponding to each swimmer is synchronously sent to the database for storage.

8. The laser radar-based swimming pool drowning video monitoring system according to claim 7, characterized in that: The acquisition unit installs a lidar sensor and an RGB camera group in the swimming pool to collect basic information data of people in the swimming pool and motion data in real time, and sends them to the lidar module and video camera module respectively. Specifically: The deep pool and shallow pool are divided into several swimming areas in equal proportions. RGB cameras are installed according to the number of swimming areas. One RGB camera collects data information of each swimmer in one swimming area and sends the acquired data information to the video camera module. A lidar sensor is installed directly above the center of the swimming pool to collect target point data of each swimmer in real time and send it to the lidar module.

Images