Anti-drowning early warning method for lakes and reservoirs

Abstract

The invention relates to the technical field of anti-drowning early warning, in particular to an anti-drowning early warning method for lakes and reservoirs, which identifies and analyzes human body behaviors in a lake and reservoir area in real time through video monitoring and deep learning technologies, provides accurate drowning early warning and danger level judgment, enhances safety protection measures, reduces the occurrence rate of drowning accidents, and improves the early warning effect of the lakes and reservoirs. And public safety is guaranteed.

Description

Technical field:

[0001] The present invention relates to the technical field of drowning prevention and early warning, and in particular to an drowning prevention and early warning method for lakes and reservoirs. Background technology:

[0002] Lake and reservoir drowning prevention warning refers to the timely identification of potential drowning hazards through monitoring and analysis of human activities in the waters, and issuing warnings to relevant personnel to reduce the occurrence of drowning accidents.

[0003] In the existing technology, lake and reservoir drowning prevention warning mainly relies on manual patrols, simple water area monitoring and limited alarm equipment. These methods are mostly regular inspections or rely on lifeguard monitoring near the water area, lacking systematic technical means. Although some areas have begun to use basic video surveillance systems, their recognition capabilities are usually limited by environmental conditions and cannot effectively distinguish between normal activities and dangerous behaviors.

[0004] In addition, waters such as lakes and reservoirs are highly open, have long coastlines and a wide range of activities, which makes it difficult to fully control human behavior and increases the risk of drowning. When people approach dangerous waters, there is a lack of timely and effective early warning mechanisms, and potential drowning accidents cannot be prevented in advance, affecting safety assurance capabilities. Traditional monitoring systems are often unable to accurately identify active objects in the waters, resulting in frequent false alarms and missed alarms. Relying on manual monitoring methods, they are often unable to respond to sudden dangerous behaviors in a timely manner, which may delay rescue opportunities. Existing systems lack the application of deep learning or intelligent analysis technology, resulting in the inability to achieve real-time analysis and intelligent judgment of behavior, and a lack of effective analysis mechanisms for behavioral data. Existing video analysis methods are mostly based on intrusion detection principles and are significantly affected by external environmental factors, especially in complex environments, which leads to a significant decrease in the accuracy of the early warning system, increases the risk of false alarms, and cannot provide a scientific basis for early warning.

[0005] In view of the shortcomings of existing lake and reservoir drowning prevention and warning technology, a more efficient and intelligent solution is urgently needed. Therefore, it is necessary to develop a drowning prevention and warning method for lakes and reservoirs. The introduction of video monitoring and deep learning technology can significantly improve the accuracy and real-time performance of monitoring. By building a data-driven behavior model, it is possible to accurately identify and warn of complex behaviors, effectively reducing the occurrence of drowning accidents. Summary of the invention:

[0006] In response to the above-mentioned defects and problems, the present invention provides a drowning prevention and warning method for lakes and reservoirs, which aims to use video monitoring and deep learning technology to identify and analyze human behavior in lake and reservoir areas in real time, provide accurate drowning warnings and danger level determinations, enhance safety protection measures, reduce the incidence of drowning accidents, and ensure public safety.

[0007] The solution adopted by the present invention to solve the technical problem is: a drowning prevention warning method for lakes and reservoirs, characterized in that it includes the following steps:

[0008] S1: Deploy a video surveillance system to record videos and identify people in the control area;

[0009] S1.1: The video monitoring system monitors and records the lake and reservoir control area, uses a video activity detection model to capture living things that enter the video screen, and uses a deep learning model to identify living things in the video screen. When the living thing is identified as a human body, the human behavior is detected and the video is continuously recorded. Otherwise, only the video is continuously recorded;

[0010] S1.2: The deep learning model automatically learns human features from the annotated data, extracts image features through the convolutional neural network, and detects the position and size of the human body in the video frame. During the detection process, the deep learning model extracts the shape and contour features of the human body, distinguishes the human body from other living things, and combines the posture estimation algorithm to identify the key points of the human body;

[0011] S2: Human behavior detection and risk level determination;

[0012] S2.1: Differentiation of control areas: distinguish the shore area and the water surface area in the video screen, use the region growing algorithm on the video frame image, calculate the shore area and the water surface area, manually or automatically select the seed points of the shore area and the water surface area, and add adjacent pixels to the corresponding area according to the similarity of gray value, color and texture between pixels, grow a complete regional division, and mark the shore area and the water surface area in the image;

[0013] S2.2: Identify the human body's motion trajectory and its behavioral characteristics, including walking, swimming and drowning; mark the five key points of the human head, upper limbs, trunk and lower limbs, namely point A on the head, points B and C on the upper limbs, points D and E on the lower limbs, and point F on the trunk;

[0014] (1) When a person walks, swims, or drowns, the video monitoring system collects data on the person's posture and the location of six key points on the body;

[0015] The video monitoring system obtains the posture characteristics of the human body when walking and the spatial position relationship of the six key points of the human body, records the distance changes of the six key points of the human body relative to the water surface, and constructs a data set model of the distance changes of the key points of the human body when walking;

[0016] The video monitoring system obtains the posture characteristics of the human body when swimming and the spatial position relationship of the six key points of the human body, records the distance changes of the six key points of the human body relative to the water surface, and constructs a data set model of the distance changes of the key points of the human body in the swimming state;

[0017] The video monitoring system obtains the posture characteristics of the human body when drowning and the spatial position relationship of the six key points of the human body, records the distance changes of the six key points of the human body relative to the water surface, and constructs a data set model of the distance changes of the key points of the human body under the drowning state;

[0018] (2) In the constructed walking state data set model, swimming state data set model, and drowning state data set model, when the human body shows the three behaviors of walking, swimming, and drowning, the video monitoring system identifies the changes in the distance between the six key points of the human body and the water surface;

[0019] When a person walks, the video monitoring system identifies that the six key points of the human body are far away from the water surface area, and the six key points of the human body are obliquely intersecting the points on the water surface. The two are not located in the same vertical space. The relative positions of the six key points of the human body are A, B, C, F, D, and E from top to bottom, and the distances of the six key points of the human body from the water surface decrease in sequence, but are always positive values;

[0020] When a person is swimming, the video monitoring system recognizes that the person has regular movements in the water. The six key points of the person and the water surface are located in the same vertical space. The distances of the six key points of the person from the water surface change as follows: the distance between point A and the water surface changes back and forth between positive and negative regularly or is positive for a long time; the distances between points B and C and the water surface change back and forth between positive and negative regularly; the distance between point F and the water surface changes back and forth between positive and negative regularly or is positive or negative for a long time;

[0021] When a person is drowning, the video monitoring system identifies that the person is not moving regularly in the water, and the six key points of the person and the water surface are located in the same vertical space, and the distances of the six key points of the person from the water surface change as follows: the distance between point A and the water surface is positive, and at the same time, the distances between points B, C, D, E, and F and the water surface are positive; or the distance between point A and the water surface is positive for a long time, and at the same time, the distances between points B, C, D, E, and F and the water surface are 0;

[0022] (3) In the constructed walking state data set model, swimming state data set model, and drowning state data set model, when the human body shows the three behaviors of walking, swimming, and drowning, the video monitoring system identifies and determines the movement amplitude of the human lower limbs;

[0023] When a person walks, the gait parameters, walking cycle, time parameters and distance parameters are used to quantify the swing amplitude of the lower limbs of the body when the person walks. When the walking speed and step frequency exceed 1.2m / s and 110steps / min, the target person is judged to be in a walking state;

[0024] When a person swims, the action cycle, action rhythm, action frequency (F), stroke times and stroke effect are used as the quantitative swing amplitude of the lower limbs of the body when performing the stroke action. Action frequency = action times / time; stroke effect = swimming distance / action times, in meters / time. The video monitoring system identifies the stroke times of the human arm per minute and calculates the action frequency. When the action frequency exceeds 60 times / 1 minute, it is determined that the target person is in a swimming state.

[0025] The distance between the upper limbs B and C and the water surface changes from negative to positive to negative, which is recorded as one paddling action. The video monitoring system records the total number of changes in the distance between the upper limbs B and C and the water surface as N times. The video recording time is t. The characteristic values ​​monitored by the video monitoring system can be used to determine whether the person is drowning. The determination process is as follows:

[0026] The formula for calculating the movement frequency of human body when swimming is:

[0027]

[0028] Among them, N is the total number of changes in the distance between the two key points B and C of the upper limbs of the human body and the water surface, and T is the time of video recording;

[0029] The formula for calculating the stroke speed of a human body when swimming is:

[0030]

[0031] Among them, X i is the position coordinate of the key point F of the human body in the i-th frame image, is the average position of the key point F of the human body in M ​​consecutive frames, t = aT, where a is an integer, a = 0, 1, 2, 3..., T is the sampling time, and t is the video playback time;

[0032] When a human body is drowning, the movement frequency and the stroke speed slow down and eventually become zero. If the movement frequency of the human body in the water is between 50 times / minute and 70 times / minute, the target human body is judged to be swimming in the water; if the movement frequency of the human body in the water is less than 50 times / minute, the target human body is judged to be drowning; at the same time, if the human body's stroke speed is between 0.42m / s and 0.6m / s, the target human body is judged to be swimming in the water; if the human body's stroke speed is less than 0.42m / s, the target human body is judged to be drowning;

[0033] The video monitoring system identifies the position information of the six key points of the human body in the detection image, records the position information, extracts the corresponding human posture features, calculates the distance values ​​between the six key points of the human body in the detection image and the water surface, determines the relationship between these distance values ​​and zero, and determines the position state of the key points. The recorded position information and the extracted posture features are matched and compared with the previously constructed human key point distance change data set model, and according to the matching and comparison results, the human motion trajectory identification results and the behavior feature recognition results are output, so as to judge and identify the behavior state of the target individual;

[0034] S2.3: Determine the degree of danger based on human behavior characteristics, and divide the control area into different degrees of danger. Set a safety warning line on the shore of the water area, so that the shore area and the water surface area are divided by the safety warning line. The danger levels include yellow warning, orange warning and red warning. If the video monitoring system identifies that a person crosses the safety warning line and enters the water surface area, the system will issue an orange or red warning. If the video monitoring system identifies that a person has not crossed the safety warning line and is always in the shore area, the system will issue a yellow warning.

[0035] If a person is walking on the shore and meets the posture characteristics of walking, the system will determine the danger level as yellow warning; if a person is swimming in the water and meets the posture characteristics of swimming, the system will determine the danger level as orange warning; if a person is drowning in the water and meets the posture characteristics of drowning, the system will determine the danger level as red warning;

[0036] S3: Voice warning and early warning response. According to different levels of danger, the corresponding voice in the voice library is matched and played in a loop in the control area. Different early warning information is sent to the background according to different levels of danger, and the video monitoring system performs real-time tracking and records the video.

[0037] Furthermore, in step (1) of S2.2, the posture characteristics of a human body when walking are that the head is straight, the body is natural, the arms are naturally hanging on both sides of the body, and the steps are rhythmic. The position information characteristics of the six key points of the human body in space when walking are: point A (head) is relatively stable or moves forward and backward, points B and C (upper limbs) move alternately on both sides above point F (trunk), point F is relatively stable or moves forward and backward, and points D and E (lower limbs) move alternately on both sides below point F;

[0038] The posture characteristics of the human body when swimming are that the head and torso float on the water surface intermittently, the arms are bent, and the body swings up and down on the water surface at a stable frequency and rhythm. The position information characteristics of the six key points of the human body in space when swimming are as follows: point A (head) appears regularly on the water surface alternately, points B and C (upper limbs) also appear regularly on the water surface alternately, and point F (trunk) appears regularly above the water surface or stays above the water surface for a long time;

[0039] The posture characteristics of a drowning person are unnatural body posture, irregular movement of the head, arms and lower limbs on the water surface, the body being below the water surface for a long time or floating on the water surface for a long time. The position information characteristics of the six key points of a drowning person in space are as follows: point A (head) appears on the water surface irregularly, while point B, C (upper limbs), D, E (lower limbs) and point F (trunk) also appear on the water surface irregularly; or point A (head) is below the water surface for a long time; or point A (head) is above the water surface for a long time, while point B, C (upper limbs), D, E (lower limbs) and point F (trunk) do not swing.

[0040] Furthermore, in step (3) of S2.2, the time parameters include single step time, stride time, step speed, and step frequency; the distance parameters include step length, stride length, step width, and foot angle.

[0041] Furthermore, in S1.1, for the video stream images captured by the video monitoring system, the R-C3D (Region Convolutional 3D Network) video activity monitoring model is used. The 3D fully convolutional network is first used to encode the video frames to generate time-series segments that may contain behaviors. The results are then classified and refined in the classification subnetwork, thereby classifying the selected regions into specific activities and quickly capturing living things in the video.

[0042] Furthermore, in S1.2, the deep learning model may adopt Faster R-CNN, YOLO (You Only LookOnce) series, and SSD (Single Shot MultiBox Detector).

[0043] Beneficial effects of the present invention: The present invention deploys a video monitoring system and combines it with a deep learning model to accurately identify living things entering the monitoring area, can effectively distinguish between the human body and other living things, reduce the false alarm rate, and thus achieve accurate monitoring of drowning risks; and can monitor and identify the three basic behaviors of walking, swimming and drowning in real time, timely judge the behavioral state of the individual, and distinguish between normal activities and potentially dangerous behaviors, provide dynamic monitoring capabilities, and help to take intervention measures in time to reduce the probability of drowning accidents; and use a regional growing algorithm to divide the shore area and the water surface area, combined with image preprocessing, to improve image resolution and target recognition accuracy; in behavioral detection, the system can identify and monitor the six key points of the human body (head, upper limbs, lower limbs and torso), and through the spatial relationship and motion trajectory of these key points, It realizes accurate analysis of behavior; and through data collection on the changes in the positions of key points in walking, swimming and drowning and the changes in the distance between them and the water surface, it builds behavioral models in different states. The system can accurately match and compare the movement characteristics of key points in different states, thereby improving the flexibility and accuracy of behavior recognition; it establishes a three-level danger warning mechanism of yellow, orange and red, so that the system can respond quickly according to the identified human behavior state, so as to take corresponding protective measures under different degrees of danger, thereby improving public safety protection capabilities; when a dangerous state is identified, the system can trigger corresponding voice warnings according to different degrees of danger and notify surrounding people in real time; through quantitative analysis of parameters such as the amplitude of movement of the lower limbs of the human body and the frequency of swimming movements, the system can more accurately judge the individual's behavior state and its safety. Description of the drawings:

[0044] Figure 1 Schematic diagram of the six key points marked on the human body.

[0045] Figure 2 Schematic diagram of the regional division between the shore area and the water surface area.

[0046] Figure 3 It is a detection flow chart of the present invention. Specific implementation method:

[0047] The present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0048] Embodiment 1, in the prior art, lake and reservoir drowning prevention warning mainly relies on manual patrols, simple water area monitoring and limited alarm equipment. These methods are mostly scheduled inspections or rely on lifeguard monitoring near the water area, lacking systematic technical means. Although some areas have begun to use basic video surveillance systems, their recognition capabilities are usually limited by environmental conditions and cannot effectively distinguish between normal activities and dangerous behaviors. Traditional monitoring systems are often unable to accurately identify active objects in the water area, resulting in frequent false alarms and missed alarms. Relying on manual monitoring, they are often unable to respond to sudden dangerous behaviors in a timely manner, which may delay rescue opportunities. The existing system lacks the application of deep learning or intelligent analysis technology, resulting in the inability to achieve real-time analysis and intelligent judgment of behavior, and also lacks an effective analysis mechanism for behavioral data. Existing video analysis methods are mostly based on intrusion detection principles and are significantly affected by external environmental factors, especially in complex environments, resulting in a significant decrease in the accuracy of the warning system, increasing the risk of false alarms, and failing to provide a scientific basis for warning.

[0049] In view of the above problems, this embodiment provides a drowning prevention warning method for lakes and reservoirs, comprising the following steps:

[0050] S1: Deploy a video surveillance system to record videos and identify people in the control area;

[0051] S1.1: The video monitoring system monitors and records the lake and reservoir control area, uses a video activity detection model to capture living things that enter the video screen, and uses a deep learning model to identify living things in the video screen. When the living thing is identified as a human body, the human behavior is detected and the video is continuously recorded, otherwise only the video is continuously recorded; for the video stream images captured by the video monitoring system, the R-C3D (Region Convolutional 3D Network) video activity monitoring model is used to first encode the video frames using a 3D fully convolutional network to generate time-series segments that may contain behaviors, and then classify and refine the results in the classification subnetwork, so as to classify the selected area into a specific activity and quickly capture the living things in the video;

[0052] S1.2: After detecting the living creature in the video, the deep learning model is used to identify the living creature. The deep learning model can use Faster R-CNN, YOLO (You Only Look Once) series, SSD (Single Shot MultiBox Detector). The deep learning model automatically learns human body features from a large amount of labeled data, extracts image features through convolutional neural networks, and detects the position and size of the human body in the video frame. At the same time, during the detection process, the deep learning model extracts the shape and contour features of the human body, distinguishes the human body from other living creatures, and combines the posture estimation algorithm to identify the key points of the human body; it helps to further verify whether the detected living creature has the structure and movement pattern of the human body. At the same time, it can also combine environmental factors such as weather, lighting, terrain and vehicle video data to optimize the deep learning model, so that the deep learning model can be better applied to other scenarios and improve the accuracy and robustness of the deep learning model;

[0053] S2: Human behavior detection and risk level determination;

[0054] S2.1: Distinguish the control area, distinguish the shore area and the water surface area in the video screen, use the regional growing algorithm for the video frame image, calculate the shore area and the water surface area, and perform grayscale, denoising and binarization image preprocessing on the video frame image to reduce the impact of noise and other irregularities in the image on the segmentation result; manually or automatically select the seed points of the shore area and the water surface area, and add the adjacent pixels to the corresponding area according to the similarity of the gray value, color and texture between the pixels to grow a complete regional division. If the number of regional divisions determined in this way is large, the similarity threshold between the pixels can be adjusted until each point in the image belongs to a region and the division is completed, thereby marking the shore area and the water surface area in the image;

[0055] S2.2: Identify the trajectory of human movement and its behavioral characteristics, including walking, swimming and drowning; e.g. Figure 1 As shown, five key points of the human head, upper limbs, trunk and lower limbs are marked, namely point A on the head, points B and C on the upper limbs, points D and E on the lower limbs, and point F on the trunk;

[0056] (1) When a person walks, swims, or drowns, the video monitoring system collects data on the person's posture and the location of six key points on the body;

[0057] The posture characteristics of a person walking are that the head is straight, the body is natural, the arms are naturally hanging on both sides of the body, and the steps are rhythmic. The position information characteristics of the six key points in the space when the person walks are as follows: point A (head) is relatively stable or moves forward and backward, points B and C (upper limbs) move alternately on both sides above point F (trunk), point F is relatively stable or moves forward and backward, and points D and E (lower limbs) move alternately on both sides below point F;

[0058] The video monitoring system obtains the posture characteristics of the human body when walking and the spatial position relationship of the six key points of the human body, records the distance changes of the six key points of the human body relative to the water surface, and constructs a data set model of the distance changes of the key points of the human body in the walking state for subsequent matching and comparison work;

[0059] The posture characteristics of the human body when swimming are that the head and torso float on the water surface intermittently, the arms are bent, and the body swings up and down on the water surface at a stable frequency and rhythm. The position information characteristics of the six key points of the human body in space when swimming are as follows: point A (head) appears regularly on the water surface alternately, points B and C (upper limbs) also appear regularly on the water surface alternately, and point F (trunk) appears regularly above the water surface or stays above the water surface for a long time;

[0060] The video monitoring system obtains the posture characteristics of the human body when swimming and the spatial position relationship of the six key points of the human body, records the distance changes of the six key points of the human body relative to the water surface, and constructs a data set model of the distance changes of the key points of the human body in the swimming state;

[0061] The posture characteristics of a human body when drowning are unnatural posture, irregular movement of the head, arms and lower limbs on the water surface, the body being below the water surface for a long time or floating on the water surface for a long time. The position information characteristics of the six key points of a human body in space when drowning are as follows: point A (head) appears on the water surface irregularly, and at the same time, point B, C (upper limbs), D, E (lower limbs) and point F (trunk) also appear on the water surface irregularly; or point A (head) is below the water surface for a long time; or point A (head) is above the water surface for a long time, and at the same time, point B, C (upper limbs), D, E (lower limbs) and point F (trunk) do not swing;

[0062] The video monitoring system obtains the posture characteristics of the human body when drowning and the spatial position relationship of the six key points of the human body, records the distance changes of the six key points of the human body relative to the water surface, and constructs a data set model of the distance changes of the key points of the human body under the drowning state;

[0063] By analyzing the posture features and key point position information of the three behavior states, the corresponding data set model is constructed for subsequent matching and comparison work to identify human movement behavior;

[0064] (2) In the constructed walking state data set model, swimming state data set model, and drowning state data set model, when the human body shows the three behaviors of walking, swimming, and drowning, the video monitoring system identifies the changes in the distance between the six key points of the human body and the water surface;

[0065] When a person walks, he is always in the shore area, and no part of his body enters below the water surface. The video monitoring system identifies that the six key points of the human body are far away from the water surface area, and the human body moves above the water surface. The six key points of the human body are obliquely intersected with the points on the water surface, and the two are not located in the same vertical space. The relative positions of the six key points of the human body are A, B, C, F, D, and E from top to bottom, and the distances of the six key points of the human body from the water surface decrease in sequence, but are always positive values;

[0066] When a person is swimming, the video monitoring system recognizes that the person has regular movements in the water. The six key points of the person and the water surface are located in the same vertical space. The distances of the six key points of the person from the water surface change as follows: the distance between point A and the water surface changes back and forth between positive and negative regularly or is positive for a long time; the distances between points B and C and the water surface change back and forth between positive and negative regularly; the distance between point F and the water surface changes back and forth between positive and negative regularly or is positive or negative for a long time;

[0067] When a person is drowning, the video monitoring system identifies that the person is not moving regularly in the water, and the six key points of the person and the water surface are located in the same vertical space, and the distances of the six key points of the person from the water surface change as follows: the distance between point A and the water surface is positive, and at the same time, the distances between points B, C, D, E, and F and the water surface are positive; or the distance between point A and the water surface is positive for a long time, and at the same time, the distances between points B, C, D, E, and F and the water surface are 0;

[0068] (3) In the constructed walking state data set model, swimming state data set model, and drowning state data set model, when the human body shows the three behaviors of walking, swimming, and drowning, the video monitoring system identifies and determines the movement amplitude of the human lower limbs;

[0069] When a person walks, gait parameters, walking cycle, time parameters and distance parameters are used as quantitative swing amplitudes of the lower limbs of the body when the person walks. Time parameters include single step time, stride time, pace, and step frequency. Distance parameters include step length, stride length, step width, and foot angle. The normal walking speed is 1.2m / s and the step frequency is 110steps / min. When the pace speed and step frequency exceed 1.2m / s and 110steps / min, the target person is judged to be in a walking state.

[0070] When a person swims, the action cycle, action rhythm, action frequency (F), stroke times, and stroke effect are used to quantify the swing amplitude of the lower limbs of the body when they make strokes. Action frequency = action times / time. The action frequency of normal swimming is 60 times / 1 minute. Stroke effect = swimming distance / action times, in meters / time.

[0071] Through the above process, the video monitoring system identifies the number of strokes per minute of the human arm and calculates the movement frequency. When the movement frequency exceeds 60 times / minute, it is determined that the target person is in a swimming state;

[0072] The distance between the upper limbs B and C and the water surface changes from negative to positive to negative, which is recorded as one paddling action. The video monitoring system records the total number of changes in the distance between the upper limbs B and C and the water surface as N times. The video recording time is t. The characteristic values ​​monitored by the video monitoring system can be used to determine whether the person is drowning. The determination process is as follows:

[0073] Movement frequency: The movement frequency of a drowning person gradually slows down and eventually returns to zero. The movement frequency calculation formula for a swimming person is:

[0074]

[0075] Among them, N is the total number of changes in the distance between the two key points B and C of the upper limbs of the human body and the water surface, and T is the time of video recording;

[0076] Swimming speed: The speed of a drowning person's stroke gradually slows down and eventually returns to zero. The formula for calculating the speed of a person's stroke when swimming is:

[0077]

[0078] Among them, X i is the position coordinate of the key point F of the human body in the i-th frame image, is the average position of the key point F of the human body in M ​​consecutive frames, t = aT, where a is an integer, a = 0, 1, 2, 3..., T is the sampling time, and t is the video playback time;

[0079] When a human body is drowning, the above-mentioned movement frequency and stroke speed slow down and eventually become zero. If the movement frequency of the human body in the water is between 50 times / minute and 70 times / minute, the target human body is judged to be swimming in the water; if the movement frequency of the human body in the water is less than 50 times / minute, the target human body is judged to be drowning; at the same time, if the human body's stroke speed is between 0.42m / s and 0.6m / s, the target human body is judged to be swimming in the water; if the human body's stroke speed is less than 0.42m / s, the target human body is judged to be drowning;

[0080] The video monitoring system identifies the position information of the six key points of the human body in the detection image, records the position information, extracts the corresponding human posture features, calculates the distance values ​​between the six key points of the human body in the detection image and the water surface, determines the relationship between these distance values ​​and zero, and determines the position state of the key points. The recorded position information and the extracted posture features are matched and compared with the previously constructed human key point distance change data set model, and according to the matching and comparison results, the human motion trajectory identification results and the behavior feature recognition results are output, so as to judge and identify the behavior state of the target individual;

[0081] S2.3: Determine the degree of danger based on human behavior characteristics, and divide the control area into different degrees of danger. Set a safety warning line on the shore of the water area, so that the shore area and the water surface area are divided by the safety warning line. The danger levels include yellow warning, orange warning and red warning. If the video monitoring system identifies that a person crosses the safety warning line and enters the water surface area, the system will issue an orange or red warning. If the video monitoring system identifies that a person has not crossed the safety warning line and is always in the shore area, the system will issue a yellow warning.

[0082] The human body is walking on the shore. It is located on the shore and is just walking on the shore. It meets the walking posture characteristics. At this time, the target will not be in danger for the time being, but the target has entered the control area and needs to be reminded to pay attention to safety. The system determines the danger level as yellow warning;

[0083] The human body is in a swimming state in the water, far away from the shore, and meets the swimming posture characteristics. At this time, the target may be in danger, but the target has entered the control area and needs to be reminded to pay attention to safety. The system determines the danger level as orange warning;

[0084] When a person is drowning in water and meets the characteristics of drowning posture, the target is in danger and needs rescue and help from staff. The system determines the danger level as a red warning;

[0085] S3: Voice warning and early warning response, matching the corresponding voice in the voice library according to different danger levels, and playing it in a loop in the control area;

[0086] Warning sounds include:

[0087] When a yellow warning occurs, the player plays "You have entered the controlled area, do not approach the water, please pay attention to safety and leave the controlled area in time!";

[0088] When an orange warning occurs, the player plays "You have entered the controlled waters. Staying in the area is not allowed. Go ashore and leave the controlled area as soon as possible to avoid accidents!";

[0089] When a red alert occurs, the player plays "There is a drowning situation in the controlled waters. The staff has been notified and is on their way to rescue!";

[0090] Different warning information is sent to the background according to different danger levels, and the video monitoring system conducts real-time tracking and records the video;

[0091] The warning information specifically includes:

[0092] When a yellow warning occurs, a text message will be sent to the backend staff, specifically "Someone has entered the control area, please continue to pay attention";

[0093] When an orange warning occurs, a voice prompt is sent to the back-end staff, specifically "Someone has entered the controlled waters, please continue to pay attention and take measures";

[0094] When a red alert occurs, text and voice prompts will be sent to backstage staff and an alarm will sound. The specific prompt will be "Someone is drowning in the controlled waters, please go to rescue immediately."

[0095] By optimizing the video monitoring system through deep learning models, it is possible to effectively distinguish between the human body and other living things, reduce the false alarm rate, and thus achieve accurate monitoring of drowning risks. The system can monitor and identify three behaviors of walking, swimming and drowning in real time, and quickly judge the behavioral state of the individual, which helps to take intervention measures in time and reduce the probability of drowning accidents. The regional growing algorithm is used to divide the shore area and the water surface area, ensuring the clarity of the monitoring image and the accuracy of the regional division, providing a reliable foundation for subsequent behavioral analysis. By constructing a data set model of walking, swimming and drowning states, the system can accurately match and compare the key point motion characteristics under different states, improving the flexibility and accuracy of behavior recognition, and establishing three-level danger warnings of yellow, orange and red. The mechanism enables the system to respond quickly according to the identified behavioral state, so as to take corresponding protective measures under different danger levels, thereby improving the public safety protection capability. Through the voice warning system, safety warnings are issued to the surrounding personnel in a timely manner, which enhances risk awareness and vigilance, promotes the public's self-protection ability, and records the video data in the monitoring process in real time, which provides a basis for subsequent event analysis and responsibility tracing, improves the efficiency and fairness of accident handling, provides a scientific basis for the quantitative analysis of human body movement amplitude and behavior frequency, enhances the reliability of behavioral state judgment, and provides stronger technical support for accident prevention. Through the combination of multiple technologies, this solution not only enhances the monitoring and early warning effect of drowning prevention, but also provides comprehensive and scientific technical support for lake and reservoir safety monitoring.

[0096] The above description is only a preferred embodiment of the present invention and does not limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A drowning prevention warning method for lakes and reservoirs, characterized in that: The following steps are involved: S1: Deploy a video surveillance system to record videos and identify people in the control area; S1.1: The video monitoring system monitors and records the lake and reservoir control area, uses a video activity detection model to capture living things that enter the video screen, and uses a deep learning model to identify living things in the video screen. When the living thing is identified as a human body, the human behavior is detected and the video is continuously recorded. Otherwise, only the video is continuously recorded; S1.2: The deep learning model automatically learns human features from the annotated data, extracts image features through the convolutional neural network, and detects the position and size of the human body in the video frame. During the detection process, the deep learning model extracts the shape and contour features of the human body, distinguishes the human body from other living things, and combines the posture estimation algorithm to identify the key points of the human body; S2: Human behavior detection and risk level determination; S2.1: Differentiation of control areas: distinguish the shore area and the water surface area in the video screen, use the region growing algorithm on the video frame image, calculate the shore area and the water surface area, manually or automatically select the seed points of the shore area and the water surface area, and add adjacent pixels to the corresponding area according to the similarity of gray value, color and texture between pixels, grow a complete regional division, and mark the shore area and the water surface area in the image; S2.2: Identify the human body's motion trajectory and its behavioral characteristics, including walking, swimming and drowning; mark the five key points of the human head, upper limbs, trunk and lower limbs, namely point A on the head, points B and C on the upper limbs, points D and E on the lower limbs, and point F on the trunk; (1) When a person walks, swims, or drowns, the video monitoring system collects data on the person's posture and the location of six key points on the body; The video monitoring system obtains the posture characteristics of the human body when walking and the spatial position relationship of the six key points of the human body, records the distance changes of the six key points of the human body relative to the water surface, and constructs a data set model of the distance changes of the key points of the human body when walking; The video monitoring system obtains the posture characteristics of the human body when swimming and the spatial position relationship of the six key points of the human body, records the distance changes of the six key points of the human body relative to the water surface, and constructs a data set model of the distance changes of the key points of the human body in the swimming state; The video monitoring system obtains the posture characteristics of the human body when drowning and the spatial position relationship of the six key points of the human body, records the distance changes of the six key points of the human body relative to the water surface, and constructs a data set model of the distance changes of the key points of the human body under the drowning state; (2) In the constructed walking state data set model, swimming state data set model, and drowning state data set model, when the human body shows the three behaviors of walking, swimming, and drowning, the video monitoring system identifies the changes in the distance between the six key points of the human body and the water surface; When a person walks, the video monitoring system identifies that the six key points of the human body are far away from the water surface area, and the six key points of the human body are obliquely intersecting the points on the water surface. The two are not located in the same vertical space. The relative positions of the six key points of the human body are A, B, C, F, D, and E from top to bottom, and the distances of the six key points of the human body from the water surface decrease in sequence, but are always positive values; When a person is swimming, the video monitoring system recognizes that the person has regular movements in the water. The six key points of the person and the water surface are located in the same vertical space. The distances of the six key points of the person from the water surface change as follows: the distance between point A and the water surface changes back and forth between positive and negative regularly or is positive for a long time; the distances between points B and C and the water surface change back and forth between positive and negative regularly; the distance between point F and the water surface changes back and forth between positive and negative regularly or is positive or negative for a long time; When a person is drowning, the video monitoring system identifies that the person is not moving regularly in the water, and the six key points of the person and the water surface are located in the same vertical space, and the distances of the six key points of the person from the water surface change as follows: the distance between point A and the water surface is positive, and at the same time, the distances between points B, C, D, E, and F and the water surface are positive; or the distance between point A and the water surface is positive for a long time, and at the same time, the distances between points B, C, D, E, and F and the water surface are 0; (3) In the constructed walking state data set model, swimming state data set model, and drowning state data set model, when the human body shows the three behaviors of walking, swimming, and drowning, the video monitoring system identifies and determines the movement amplitude of the human lower limbs; When a person walks, the gait parameters, walking cycle, time parameters and distance parameters are used to quantify the swing amplitude of the lower limbs of the body when the person walks. When the walking speed and step frequency exceed 1.2m / s and 110steps / min, the target person is judged to be in a walking state; When a person swims, the action cycle, action rhythm, action frequency (F), stroke times and stroke effect are used as the quantitative swing amplitude of the lower limbs of the body when performing the stroke action. Action frequency = action times / time; stroke effect = swimming distance / action times, in meters / time. The video monitoring system identifies the number of strokes per minute of the human arm and calculates the action frequency. When the action frequency exceeds 60 times / 1 minute, it is determined that the target person is in a swimming state. The distance between the upper limbs B and C and the water surface changes from negative to positive to negative, which is recorded as one paddling action. The video monitoring system records the total number of changes in the distance between the upper limbs B and C and the water surface as N times. The video recording time is t. The characteristic values ​​monitored by the video monitoring system can be used to determine whether the person is drowning. The determination process is as follows: The formula for calculating the movement frequency of human body when swimming is: Among them, N is the total number of changes in the distance between the two key points B and C of the upper limbs of the human body and the water surface, and T is the time of video recording; The formula for calculating the stroke speed of a human body when swimming is: Among them, X i is the position coordinate of the key point F of the human body in the i-th frame image, is the average position of the key point F of the human body in M ​​consecutive frames, t = aT, where a is an integer, a = 0, 1, 2, 3..., T is the sampling time, and t is the video playback time; When a human body is drowning, the movement frequency and the stroke speed slow down and eventually become zero. If the movement frequency of the human body in the water is between 50 times / minute and 70 times / minute, the target human body is judged to be swimming in the water; if the movement frequency of the human body in the water is less than 50 times / minute, the target human body is judged to be drowning; at the same time, if the human body's stroke speed is between 0.42m / s and 0.6m / s, the target human body is judged to be swimming in the water; if the human body's stroke speed is less than 0.42m / s, the target human body is judged to be drowning; The video monitoring system identifies the position information of the six key points of the human body in the detection image, records the position information, extracts the corresponding human posture features, calculates the distance values ​​between the six key points of the human body in the detection image and the water surface, determines the relationship between these distance values ​​and zero, and determines the position state of the key points. The recorded position information and the extracted posture features are matched and compared with the previously constructed human key point distance change data set model, and according to the matching and comparison results, the human motion trajectory identification results and the behavior feature recognition results are output, so as to judge and identify the behavior state of the target individual; S2.3: Determine the degree of danger based on human behavior characteristics, and divide the control area into different degrees of danger. Set a safety warning line on the shore of the water area, so that the shore area and the water surface area are divided by the safety warning line. The danger levels include yellow warning, orange warning and red warning. If the video monitoring system identifies that a person crosses the safety warning line and enters the water surface area, the system will issue an orange or red warning. If the video monitoring system identifies that a person has not crossed the safety warning line and is always in the shore area, the system will issue a yellow warning. If a person is walking on the shore and meets the posture characteristics of walking, the system will determine the danger level as yellow warning; if a person is swimming in the water and meets the posture characteristics of swimming, the system will determine the danger level as orange warning; if a person is drowning in the water and meets the posture characteristics of drowning, the system will determine the danger level as red warning; S3: Voice warning and early warning response. According to different levels of danger, the corresponding voice in the voice library is matched and played in a loop in the control area. Different early warning information is sent to the background according to different levels of danger, and the video monitoring system performs real-time tracking and records the video.

2. The drowning prevention warning method for lakes and reservoirs according to claim 1, characterized in that: In step (1) of S2.2, the posture characteristics of a person walking are that the head is straight, the body is natural, the arms are naturally hanging on both sides of the body, and the steps are rhythmic. The position information characteristics of the six key points of the person walking in space are: point A (head) is relatively stable or moves forward and backward, points B and C (upper limbs) move alternately on both sides above point F (trunk), point F is relatively stable or moves forward and backward, and points D and E (lower limbs) move alternately on both sides below point F; The posture characteristics of the human body when swimming are that the head and torso float on the water surface intermittently, the arms are bent, and the body swings up and down on the water surface at a stable frequency and rhythm. The position information characteristics of the six key points of the human body in space when swimming are as follows: point A (head) appears regularly on the water surface alternately, points B and C (upper limbs) also appear regularly on the water surface alternately, and point F (trunk) appears regularly above the water surface or stays above the water surface for a long time; The posture characteristics of a drowning person are unnatural body posture, irregular movement of the head, arms and lower limbs on the water surface, the body being below the water surface for a long time or floating on the water surface for a long time. The position information characteristics of the six key points of a drowning person in space are as follows: point A (head) appears on the water surface irregularly, while point B, C (upper limbs), D, E (lower limbs) and point F (trunk) also appear on the water surface irregularly; or point A (head) is below the water surface for a long time; or point A (head) is above the water surface for a long time, while point B, C (upper limbs), D, E (lower limbs) and point F (trunk) do not swing.

3. The drowning prevention warning method for lakes and reservoirs according to claim 1, characterized in that: In step (3) of S2.2, the time parameters include single step time, stride time, step speed, and step frequency; the distance parameters include step length, stride length, step width, and foot angle.

4. The drowning prevention warning method for lakes and reservoirs according to claim 1, characterized in that: In S1.1, for the video stream images captured by the video monitoring system, the R-C3D (Region Convolutional 3D Network) video activity monitoring model is used. The 3D fully convolutional network is first used to encode the video frames to generate time-series segments that may contain behaviors. The classification subnetwork then classifies and refines the results, thereby classifying the selected regions into specific activities and quickly capturing living things in the video.

5. The drowning prevention warning method for lakes and reservoirs according to claim 1, characterized in that: In S1.2, the deep learning model can use Faster R-CNN, YOLO (You Only Look Once) series, and SSD (Single Shot MultiBox Detector).

Images