Scenic spot potential safety hazard inspection method and device, electronic equipment and medium

Abstract

The embodiment of the invention discloses a scenic spot potential safety hazard inspection method and device, electronic equipment and a storage medium. The method comprises the steps that live broadcast video streams of live broadcast point locations of a target scenic spot are acquired; identifying hidden danger information corresponding to each live broadcast point location based on a preset identification model and each live broadcast video stream; performing time sequence analysis on the identified hidden danger information; and generating a risk level corresponding to the hidden danger information according to the historical inspection data of the target scenic area and a time sequence analysis result, and performing safety inspection on each live broadcast point location according to the risk level. According to the scheme, the safety prevention and control efficiency of the scenic area can be improved, and meanwhile, the safety prevention and control reliability is improved.

Description

Technical Field

[0001] This application relates to the field of security monitoring technology, specifically to a method, device, electronic equipment, and storage medium for inspecting safety hazards in scenic areas. Background Technology

[0002] In the field of scenic area safety management, the demand for comprehensive and real-time hazard monitoring and multi-risk linkage prevention and control is growing. Currently, existing traditional scenic area safety inspection methods have certain limitations. Traditional inspection methods rely heavily on manual patrols and simple equipment monitoring, making it difficult to accurately address complex safety scenarios. For example, if management wants to monitor the "linkage risk between the expansion of cracks in boardwalks during the rainy season and congestion in the surrounding area," traditional inspections may only record the crack condition and visitor flow data separately, failing to correlate and analyze the combined safety hazards. Furthermore, these inspection methods struggle to effectively correlate and integrate information from various points, such as facility status, environmental factors, and visitor behavior, leading to the need for repeated inspections to grasp the overall risk, resulting in low safety control efficiency and reliability. Summary of the Invention

[0003] This application provides a method, electronic device, apparatus, and storage medium for inspecting safety hazards in scenic areas based on live streaming, which can improve the efficiency and reliability of safety control in scenic areas.

[0004] In a first aspect, embodiments of this application provide a method for inspecting safety hazards in scenic areas based on live streaming, including:

[0005] Obtain live video streams from various live streaming locations within the target scenic area;

[0006] Based on the preset recognition model and each of the live video streams, identify the hidden danger information corresponding to each live broadcast location;

[0007] Perform time-series analysis on the identified hazard information;

[0008] Based on the historical inspection data and time-series analysis results of the target scenic area, a risk level corresponding to the hidden danger information is generated, and safety inspections are carried out on each live broadcast location according to the risk level.

[0009] Optionally, in some embodiments of this application, the step of identifying the hidden danger information corresponding to each live broadcast location based on a preset identification model and each of the live video streams includes:

[0010] Obtain the preset recognition model;

[0011] Extract video frames from the live video streams of each live streaming location;

[0012] The captured video frames are input into the recognition model, which outputs the hazard information corresponding to each live broadcast location.

[0013] Optionally, in some embodiments of this application, the step of extracting video frames from the live video stream of each live broadcast location includes:

[0014] Video frames are extracted from the live video stream of each live location according to the preset frame rate.

[0015] The captured video frames are subjected to image enhancement processing to obtain the processed video frames.

[0016] Optionally, in some embodiments of this application, the step of performing time-series analysis on the identified hazard information includes:

[0017] Based on the potential hazard information, consecutive video frames are selected from the live video stream as analysis samples;

[0018] The analysis samples are subjected to time-series analysis based on the time-series dimension.

[0019] Optionally, in some embodiments of this application, the step of performing time-series analysis on the analysis samples based on the time-series dimension includes:

[0020] Calculate the rate of change of passenger flow density of the analyzed sample per unit time, and output the passenger flow risk corresponding to the analyzed sample based on the rate of change of passenger flow density, and / or;

[0021] Monitor the defect changes of the first target object in the analysis sample within a unit time period, and output the object defect risk corresponding to the analysis sample based on the defect changes of the target object, and / or;

[0022] The pixel percentage of the second target object in each image frame of the analysis sample is monitored, and the object spread risk corresponding to the analysis sample is output according to the increase of the second target object in a unit time.

[0023] Optionally, in some embodiments of this application, the step of generating a risk level corresponding to the hidden danger information based on the historical inspection data and time-series analysis results of the target scenic area, and conducting safety inspections of each live broadcast location based on the risk level, includes:

[0024] Historical risk information corresponding to each live broadcast location is obtained from the historical inspection data of the target scenic area.

[0025] Based on the historical risk information and time series analysis results, output the fused data corresponding to each live broadcast location;

[0026] Based on the fused data, a risk level corresponding to the hidden danger information is generated;

[0027] Generate inspection strategies corresponding to the risk levels, and conduct security inspections on each live broadcast location based on the inspection strategies.

[0028] Optionally, in some embodiments of this application, generating the risk level corresponding to the hazard information based on the fused data includes:

[0029] Determine the degree of risk matching corresponding to the fused data;

[0030] Based on the risk matching degree and the preset offset range, the risk level corresponding to the hidden danger information is generated.

[0031] Optionally, in some embodiments of this application, generating the inspection strategy corresponding to the risk level includes:

[0032] Determine the target audience, push frequency, and processing information corresponding to each risk level;

[0033] Based on the push target, push frequency, and processing information, an inspection strategy corresponding to the risk level is generated.

[0034] Secondly, embodiments of this application provide a scenic area safety hazard inspection device based on live streaming of scenic areas, comprising:

[0035] The acquisition module is used to acquire live video streams from various live streaming locations within the target scenic area.

[0036] The identification module is used to identify potential hazard information corresponding to each live broadcast location based on a preset identification model and each of the live video streams.

[0037] The analysis module is used to perform time-series analysis on the identified hazard information;

[0038] The generation module is used to generate the risk level corresponding to the hidden danger information based on the historical inspection data and time series analysis results of the target scenic area.

[0039] The inspection module is used to conduct safety inspections of each live broadcast location based on the risk level.

[0040] Accordingly, this application also provides an electronic device, including a memory, a processor, and a processor program stored in the memory and executable on the processor, wherein the processor executes the program as described in any of the methods above.

[0041] This application also provides a storage medium storing a processor program that, when executed by a processor, implements any of the methods described above.

[0042] This application provides a method, device, electronic device, and storage medium for inspecting safety hazards in scenic areas. After acquiring live video streams from various live-streaming locations within a target scenic area, the method identifies hazard information corresponding to each live-streaming location based on a preset recognition model and the live video streams. Then, it performs time-series analysis on the identified hazard information. Finally, based on historical inspection data of the target scenic area and the time-series analysis results, it generates a risk level corresponding to the hazard information and conducts safety inspections on each live-streaming location according to the risk level. In the scenic area safety hazard inspection scheme based on live-streaming provided in this application, a preset recognition model is used to identify hazard information at each live-streaming location, and combined with historical inspection data of the target scenic area, the risk corresponding to the hazard information is determined. This allows for flexible formulation of inspection strategies, and inspections of different live-streaming locations can be conducted based on these strategies. Therefore, this scheme not only improves the efficiency of safety control in scenic areas but also enhances the reliability of safety control. Attached Figure Description

[0043] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 This is a flowchart illustrating the scenic area safety hazard inspection method based on live streaming provided in this application embodiment;

[0045] Figure 2 This is a schematic diagram of the structure of the scenic area safety hazard inspection device based on live streaming of scenic areas provided in this application embodiment;

[0046] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0047] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0048] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, components, features, and elements with the same names in different embodiments of this application may have the same meaning or different meanings, the specific meaning of which must be determined by its interpretation in that specific embodiment or further in conjunction with the context of that specific embodiment.

[0049] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0050] In the following description, the use of suffixes such as "module," "part," or "unit" to denote elements is solely for the purpose of illustrative purposes and has no specific meaning in itself. Therefore, "module," "part," or "unit" may be used interchangeably.

[0051] The following describes the embodiments involved in this application in detail. It should be noted that the order of description of the embodiments in this application is not intended to limit the priority of the embodiments.

[0052] This application provides a method, apparatus, storage medium, and smart terminal for inspecting safety hazards in scenic areas. Specifically, the scenic area safety hazard inspection method based on live streaming of scenic areas can be executed by a smart terminal or a server. The smart terminal can be a smartphone, tablet, laptop, touchscreen, game console, personal computer (PC), personal digital assistant (PDA), or other smart terminal. The terminal may also include a client, which can be a media playback client or an instant messaging client, etc.

[0053] This application provides a method for inspecting safety hazards in scenic areas, which can be executed by an electronic device or a server. This application example illustrates the method using an electronic device. The electronic device includes a touchscreen display and a processor. The touchscreen display presents a graphical user interface (GUI) and receives user commands applied to the GUI. When the user operates the GUI via the touchscreen, the GUI can control local content on the electronic device or control content on the server side in response to the received commands.

[0054] The scenic area safety hazard inspection solution provided in this application uses a preset identification model to identify hazard information at each live broadcast location and combines it with historical inspection data of the target scenic area to determine the risk corresponding to the hazard information. This allows for flexible formulation of inspection strategies, and inspections can be carried out on different live broadcast locations based on these strategies. Therefore, this solution can not only improve the efficiency of safety control in scenic areas, but also enhance the reliability of safety control.

[0055] The following sections provide detailed descriptions of each example. It should be noted that the order in which the embodiments are described is not intended to limit the priority of the embodiments.

[0056] A method for inspecting safety hazards in scenic areas includes: acquiring live video streams from various live streaming locations in the target scenic area; identifying hazard information corresponding to each live streaming location based on a preset identification model and each live video stream; performing time-series analysis on the identified hazard information; generating risk levels corresponding to the hazard information based on historical inspection data and time-series analysis results of the target scenic area; and conducting safety inspections on each live streaming location based on the risk levels.

[0057] Please see Figure 1 , Figure 1 This application provides a flowchart illustrating a method for inspecting safety hazards in scenic areas based on live streaming. The specific process of this method is as follows:

[0058] 101. Obtain the live video streams from each live streaming location in the target scenic area.

[0059] It should be noted that the live streaming locations for the target scenic area need to be planned in advance based on safety management requirements. These locations typically cover key safety hazard areas within the scenic area, such as crowded viewing platforms, boardwalks prone to aging facilities, docks near water, and areas operating special equipment (such as cable cars and cruise ships). The placement of these locations can be combined with the scenic area's terrain, facility distribution, and historical risk data to ensure that the video stream covers key monitoring scenarios (such as facility damage, tourist violations, and environmental anomalies).

[0060] In addition, the source of live video streams is mainly based on dedicated equipment deployed in scenic areas or existing monitoring systems. If the scenic area has deployed security monitoring cameras (such as high-definition network cameras), the real-time video stream can be obtained directly through the back-end platform of the monitoring system (such as NVR, video integrated management platform). Data extraction is usually performed using standard protocols such as RTSP (Real-Time Streaming Protocol) and RTMP (Real-Time Messaging Protocol). For locations that require dedicated live streaming (such as key scenic spots and high-risk areas), live streaming equipment with real-time streaming capabilities (such as 4G / 5G high-definition live streaming cameras) can be deployed. The video stream is uploaded to the scenic area's private cloud or designated server in real time through the built-in network module of the equipment (or an external 5G module), and then the corresponding live video stream is obtained from the server.

[0061] 102. Based on the preset recognition model and each live video stream, identify the hidden danger information corresponding to each live broadcast location.

[0062] The preset recognition model refers to a multi-target detection and classification model (such as YOLOv8, Faster R-CNN, etc.) trained based on deep learning technology. Its core is to be trained on a large number of scenic area safety hazard samples (such as crack images, congestion scene videos, facility damage photos, etc.) to have the ability to automatically identify specific safety risk targets.

[0063] The input to this recognition model is image frames extracted from a live video stream, and the output is the location (e.g., bounding box coordinates), type (e.g., "boardwalk crack" or "crowd congestion") and quantitative features (e.g., crack length and crowd density) of the potential hazards in the image. It can accurately match common types of safety hazards in scenic areas (e.g., facility damage, crowd gathering, environmental risks, etc.).

[0064] Hazard information refers to structured data related to scenic area safety extracted from live video streams through identification models. Specifically, it may include hazard types (such as loose guardrails, excessive crowd density), location information (camera locations and specific coordinates of the images), and timestamps, etc.

[0065] For example, a pre-trained, pre-defined recognition model (such as a YOLOv8-based multi-object recognition model) is invoked. Pre-processed image frames are input into the model for inference. Then, a model weight file containing pre-defined hazard labels such as "boardwalk cracks," "crowd congestion," "facility damage," and "abnormal behavior" is loaded to ensure the model's ability to identify specific hazard types in the target scenic area. Next, consecutive image frames from the same live-streaming location are input into the model in batches to improve recognition efficiency. During processing, the recognition model identifies targets belonging to the pre-defined hazard types in the image through feature extraction and matching, and outputs the classification results (i.e., hazard information).

[0066] Optionally, in some embodiments of this application, the step "identifying the hidden danger information corresponding to each live broadcast location based on a preset identification model and each of the live video streams" may specifically include:

[0067] Obtain the preset recognition model;

[0068] Extract video frames from the live video streams of each live streaming location;

[0069] The captured video frames are input into the recognition model, which outputs the hazard information corresponding to each live broadcast location.

[0070] The preset recognition model is a deep learning model (such as YOLOv8, Faster R-CNN and other object detection models) that is customized and trained for the types of safety hazards in the target scenic area. Its training samples include historical hazard cases of the scenic area (such as images of boardwalk cracks, video frames of crowd congestion, photos of damaged facilities, etc.) and are labeled with exclusive hazard tags such as "cracks", "congestion" and "loose guardrails".

[0071] The model is loaded through the model deployment module of the scenic area safety management system. The model file (including weight parameters and network structure) is stored on a local server or in the cloud, supporting real-time access (such as loading it into the video analysis module via API interface). This ensures that the model can accurately identify specific hazards in the scenic area (such as "cracks in the Taochuan Academy plank road in Changde Taohuayuan Scenic Area" and "congestion in the densely populated area of ​​Qin Valley"). For the real-time video streams from various live broadcast points (such as cameras covering the plank road, viewing platform, and entrance plaza), video frames are extracted according to the principle of "balancing recognition accuracy and efficiency." The specific operation is as follows: the interval is set according to the dynamic characteristics of the hazard (such as capturing 1 frame every 5 seconds for slow-changing hazards such as "crack expansion" and 2 frames per second for fast-changing hazards such as "congestion") to ensure that key nodes of hazard change are captured (such as the process of congestion from formation to deterioration).

[0072] Then, raw image frames (usually in JPG or PNG format) are extracted from the live video stream via RTSP (Real-Time Streaming Protocol) or the SDK interface of the scenic area monitoring system. For example, for a live stream with camera ID "AGX0220044DEB", the binary data of each frame is obtained in real time through protocol parsing. The extracted video frames are then standardized, including resizing (e.g., uniformly set to 640×640 pixels to match model input requirements), removing noise (e.g., raindrop interference in rainy scenes, light spots at night), and correcting colors (e.g., brightness compensation in backlit scenes) to prevent image quality from affecting the model's recognition accuracy.

[0073] Finally, the preprocessed video frames are input in batches into the preset recognition model, and the model outputs the results through the following logic:

[0074] Feature extraction: The model extracts key features from video frames (such as the edge contours of cracks, the dense distribution texture of crowds, and the structural shape of guardrails) through convolutional layers.

[0075] Target matching and classification: The extracted features are compared with the hazard label library during model training to locate the hazard location (e.g., mark "crack in the upper left corner of the image" with a bounding box), and output the type and quantitative information (e.g., "crack length 3.2cm", "passenger flow density 1.8 people / ㎡", "guardrail tilt angle 10°").

[0076] The results of this identification model are linked to the live broadcast location to generate structured hazard information, including:

[0077] Types of potential hazards (e.g., "cracks in the boardwalk" or "crowd congestion");

[0078] The camera ID and screen coordinates corresponding to the live broadcast location (e.g., “AGX0220044DEB, image coordinates (x1,y1)-(x2,y2)”);

[0079] Quantitative features (such as crack width and congestion area) and timestamps (accurate to the moment the frame was captured).

[0080] Optionally, in some embodiments of this application, the step of "extracting video frames from the live video stream of each live location" may specifically include:

[0081] Video frames are extracted from the live video stream of each live location according to the preset frame rate.

[0082] The captured video frames are subjected to image enhancement processing to obtain the processed video frames.

[0083] The preset frame rate refers to the number of video frames captured per second (unit: frames / second, fps) pre-set based on the type of hazard (such as the speed of dynamic change) at different live streaming points in the target scenic area. Its core function is to balance "recognition accuracy" and "data efficiency": for hazards that change rapidly (such as crowd congestion or sudden crowd gatherings), a higher frame rate (such as 2-3 fps) should be set to ensure that every change is captured; for hazards that change slowly (such as the expansion of cracks in boardwalks or the slow aging of facilities), a lower frame rate (such as 0.2-1 fps) can be set to avoid redundant data.

[0084] For example, based on the core monitoring hazard types of each live streaming location, a preset frame rate configuration table is invoked. Then, video frames are extracted based on the live streaming protocol. Finally, metadata is bound to each captured image frame. Image enhancement can include noise reduction, contrast and brightness optimization, and size normalization, among other things.

[0085] 103. Conduct time-series analysis on the identified hazard information.

[0086] Time series analysis refers to the analytical process of uncovering the patterns, trends, periodicity, and correlations of hazard changes over time based on a time-stamped sequence of hazard information (i.e., hazard data recorded in chronological order). Its purpose is to reveal the dynamic evolution characteristics of hazards (such as whether they continuously worsen or occur periodically), predict future development trends (when they might escalate into serious risks), and provide a basis for hazard classification, handling, and early intervention.

[0087] For example, hazard information identified at different times from various live-streaming locations can be aggregated and integrated by location-time-hazard attribute association. Hazard attributes include: hazard type (e.g., tourists crossing boundaries; loose facilities), hazard level (e.g., minor / moderate / serious), occurrence timestamp (accurate to the minute, e.g., 2025-07-31 09:15:00), and duration (e.g., tourists crossing boundaries from 09:15 to 09:18). Then, treating each live-streaming location as an independent unit, its hazard information is sorted chronologically to form a hazard time series for that location.

[0088] Blank data for which no potential hazards were identified within a certain time period can be marked as "no potential hazards".

[0089] Optionally, in some embodiments of this application, false hazards caused by misjudgment by the identification model (such as "facility cracks" misjudged due to light interference) can be eliminated by manual review or rule verification. For example, a hazard may only occur once and last for less than 10 seconds, and there may be no related hazards before or after the hazard.

[0090] Next, key features reflecting dynamic changes are extracted from the time series of potential hazards, including: frequency features, i.e., the number of times a hazard occurs per unit time; duration features, i.e., the duration of a single hazard (e.g., "loose guardrail" lasts 45 minutes from discovery to disappearance) and the total duration of consecutive occurrences (e.g., "tourist crossing the boundary at point C lasts a cumulative 20 minutes within 1 hour"); interval features, i.e., the time interval between two similar hazards (e.g., "tourist crossing the boundary at point A occurs on average once every 15 minutes") and the time difference between the first and last occurrences (reflecting the duration of the hazard); and rate of change features, i.e., the speed at which the hazard level is upgraded / downgraded (e.g., "it only takes 8 minutes for 'minor congestion' to upgrade to 'severe congestion'") and the speed at which the hazard spreads (e.g., "it takes 12 minutes for the hazard to spread from point D to the adjacent point E").

[0091] Furthermore, based on the extracted time-series features, methods such as moving average and linear regression are used to determine the overall trend of hazard changes. Simultaneously, statistical tests (such as t-tests) can be used to identify anomalous change points (abrupt changes) in the hazard sequence. In addition, based on historical time-series data, time series models (such as ARIMA and LSTM) can be used to predict the probability or level of hazard occurrence in the future. Finally, the analysis conclusions are compiled into a structured report, including trend labels for each hazard location (e.g., continuous deterioration, periodic high incidence, or occasional abrupt changes); key time nodes (e.g., daily peak risk periods, holiday hazard outbreak windows); associated hazard combinations (e.g., a chain of risks such as vehicle congestion - tourist gathering - facility overload); and predictive warning information (e.g., a 75% probability of a trestle crack expanding at location H within the next hour).

[0092] Optionally, in some embodiments of this application, the step "performing time-series analysis on the identified hazard information" may specifically include:

[0093] Based on the potential hazard information, consecutive video frames are selected from the live video stream as analysis samples;

[0094] Time series analysis is performed on the analysis samples based on the time series dimension.

[0095] The analysis sample refers to a set of continuous video frames selected from the live video stream that are directly related to specific hazard information. The temporal dimension refers to the analysis perspective centered on the chronological order of time, focusing on the temporal correlation of video frames in the analysis sample, and connecting the content changes of video frames through a timeline to explore the evolution pattern of hazards over time (such as whether they expand, move, or are accompanied by other anomalies).

[0096] Specifically, the target hazard information to be analyzed is extracted from the identification results. Then, based on the timestamp of the target hazard information, continuous video frames are extracted from the live video stream to form an analysis sample. For example, if the time range of the target hazard information is t1 to t2, continuous video is extracted from the video stream of the corresponding live location. Then, the dynamic changes of the continuous frames are analyzed from the perspective of the time axis, and the spatial position movement of the hazard subject (such as congested crowds, abnormal equipment, etc.) in the continuous frames is analyzed through inter-frame target detection.

[0097] Optionally, in some embodiments of this application, the step "performing time-series analysis on the analysis sample based on the time-series dimension" may specifically include:

[0098] Calculate the rate of change of passenger flow density for the analyzed sample per unit time, and output the passenger flow risk corresponding to the analyzed sample based on the rate of change of passenger flow density, and / or;

[0099] Monitor the defect changes of the first target object in the analysis sample within a unit of time, and output the defect risk of the object corresponding to the analysis sample based on the defect changes of the target object, and / or;

[0100] The system monitors and analyzes the pixel percentage of the second target object in each image frame and outputs the object spread risk corresponding to the analysis sample based on the increase of the second target object within a unit of time.

[0101] The rate of change of passenger flow density refers to the change in the number of tourists per unit area in the analysis sample per unit time (e.g., every 1 minute, every 5 minutes). The calculation formula is: (current passenger flow density - passenger flow density at the previous moment) / time interval. It is used to reflect the speed of passenger flow gathering or dispersal (positive value is growth, negative value is decrease).

[0102] The primary target objects refer to the fixed facilities within the scenic area that need to be monitored (such as boardwalks, railings, viewing platforms, fire-fighting equipment, etc.), whose condition directly affects the safety of tourists.

[0103] The defect change of the first target object refers to the dynamic evolution of the damaged or abnormal state of the first target object over time (such as the increase in the length of the crack in the walkway, the widening of the loose angle of the guardrail, and the increase in the displacement distance of the fire-fighting equipment).

[0104] The second target refers to dynamic risk sources such as fires or smoke that require monitoring of their spread.

[0105] The pixel percentage of the second target object in each image frame refers to the proportion of the number of pixels of the second target object in a single frame to the total number of pixels in that frame. It is used to quantify its spatial range (the higher the proportion, the larger the range). For example, the proportion of the number of pixels occupied by fire or smoke in a single frame of video image to the total number of pixels in that frame reflects its coverage area in the current scene.

[0106] The increase in the percentage of pixels of the second target object per unit time refers to the increase in the percentage of pixels of the second target object per unit time (e.g., per minute). For example, the relative increase in the percentage of pixels of fire or smoke per unit time reflects its diffusion speed.

[0107] Passenger flow risk is determined based on the rate of change in passenger flow density, reflecting whether passenger flow is rapidly gathering (easily causing congestion and stampedes) or rapidly dispersing (potentially implying sudden anomalies).

[0108] Object defect risk is a risk determined based on the defect changes of the primary target object, reflecting whether the damage to the facility continues to worsen (e.g., accelerated crack propagation may lead to collapse).

[0109] The risk of object spread is determined based on the increase in the pixel ratio of the second target object, reflecting whether the dynamic risk source (such as fire or smoke) is rapidly expanding its range (if the increase is too high, it may lead to the rapid spread of fire / smoke and expand the scope of impact).

[0110] To address visitor flow risks, the system analyzes samples (continuous video frames) using a pre-defined crowd detection and counting model (such as YOLOv8 combined with the DeepSORT algorithm) to identify the number and distribution area of ​​visitors in each frame. Combined with the actual physical area of ​​the live stream location (e.g., the entrance plaza area is 200㎡), the visitor density for each frame is calculated: Visitor density = Number of visitors per frame ÷ Actual area of ​​the location (unit: people / ㎡). Then, record the passenger flow density at key time points in chronological order. For example, t0 = 9:00: 1.0 people / ㎡, t1 = 9:02: 1.5 people / ㎡, t2 = 9:04: 2.2 people / ㎡. Next, calculate the rate of change according to the set unit time interval (e.g., 2 minutes): Passenger flow density change rate = (current passenger flow density - previous passenger flow density) ÷ time interval. For example, the change rate of t0-t1 = (1.5-1.0) ÷ 2 = 0.25 people / (㎡·minute). Finally, the preset passenger flow risk threshold table can be called, as follows:

[0111] Low risk: Change rate < 0.2 people / (㎡·minute) (slow passenger flow growth, no risk of congestion);

[0112] Medium risk: 0.2 ≤ change rate < 0.3 people / (㎡·minute) (accelerated gathering of passengers, requires attention);

[0113] High risk: Change rate ≥ 0.3 people / (㎡·minute) (rapid gathering of passengers may cause congestion and stampedes).

[0114] Therefore, the t0-t1 period can be identified as medium risk.

[0115] To address the risk of object defects, the first target object (e.g., "boardwalk railing" or "viewing platform floor tiles") is identified from the analyzed samples using a target detection model (e.g., Faster R-CNN), and its initial defects are marked (e.g., "railing joint gap width 0.3cm" or "floor tile loosening angle 5°"). Then, defects in consecutive video frames can be dynamically tracked, and changes can be quantified using image measurement tools (e.g., edge detection algorithms). For example, at t0 = 10:00, the defect feature is recorded as a 3cm railing gap, and at t1 = 10:10, the defect feature is recorded as a 7cm railing gap. Next, the change in defect per unit time (10 minutes) is calculated: Thus, the change in railing gap = 7 = 4cm, and the rate of change = 4 ÷ 10 = 0.4cm / minute.

[0116] Finally, the preset defect risk threshold table is invoked, as follows:

[0117] Low risk: Change rate < 0.2 cm / min (gap) (defect deterioration is slow);

[0118] Medium risk: 0.2 ≤ change rate < 0.5 cm / min (gap) (defects deteriorate rapidly, requiring repair);

[0119] High risk: Change rate ≥ 0.5 cm / min (gap) (defects deteriorate rapidly, which may lead to collapse and detachment).

[0120] As can be seen from the above, a guardrail gap change rate of 0.5 cm / minute is classified as "medium-level object defect risk".

[0121] To address the risk of object spread, a second target object (such as "fire," "smoke," or "waterlogged area") is identified from the analyzed samples using an image segmentation model (such as U-Net), and its pixel range in each frame is segmented. Pixel percentage = (number of pixels of the second target object ÷ total number of pixels in the frame) × 100%. Then, the increase in pixel percentage per unit time is calculated: pixel percentage increase = (current time percentage - previous time percentage) ÷ time interval.

[0122] Finally, the preset propagation risk threshold table is called to output the propagation risk of the object. For details, please refer to the previous embodiments, which will not be repeated here.

[0123] 104. Based on the historical inspection data and time-series analysis results of the target scenic area, generate the risk level corresponding to the hidden danger information, and conduct safety inspections of each live broadcast point according to the risk level.

[0124] Historical inspection data refers to various data recorded manually or systematically during past safety inspections of the target scenic area, including the type of hazard, time of occurrence, location, handling measures, impact, and historical risk results of similar hazards. Time-series analysis results refer to dynamic data on hazard changes over time obtained through time-series analysis of continuous video frames (analysis samples), such as the rate of change in visitor density per minute, the increase in the size / number of equipment defects over time, and the increase in the percentage of pixels of fire / smoke per unit time. Risk level: Hazards are classified into levels (e.g., low, medium, high) based on their severity, development speed, and potential impact, used to quantify their urgency and severity. Safety inspection refers to on-site inspection activities conducted to identify safety hazards within the scenic area and prevent accidents. Its content, frequency, and priority are determined by the risk level of the hazard (generally, high-risk locations are inspected first, and low-risk locations are inspected later).

[0125] Specifically, historical inspection data of the target scenic area can be collected, including information such as the types of hazards recorded in past inspections (e.g., equipment failure, crowd congestion, fire hazards), the time of occurrence, location, handling results, and scope of impact. Simultaneously, the time-series analysis results for current hazards can be summarized (i.e., dynamic change data of hazards obtained through continuous video frame analysis, such as the rate of change in crowd density over time, the rate of deterioration of equipment defects, and the spread trend of fire). Then, based on historical inspection data, the correlation between hazards and risks is extracted. Combined with dynamic indicators from the time-series analysis results (e.g., the rate of change in current crowd density, the rate of change in equipment defects, and the increase in the percentage of fire smoke pixels), risk level assessment rules are determined (e.g., the higher the indicator, the higher the risk level; when the indicator exceeds the historical high-risk threshold, it is directly judged as high-risk). Next, the time-series analysis results of the current hazards are substituted into the risk level assessment model and compared with similar hazard cases from historical inspection data. Output the risk level (usually divided into three levels: low, medium, and high, or more detailed classifications). For example, if the current passenger flow density change rate is 60% / minute, and historical data shows that when this indicator exceeds 50%, it has caused congestion, then it is judged as "high risk".

[0126] Finally, for each live broadcast point within the scenic area (i.e., the area covered by video surveillance), they are sorted according to the level of potential risks: high-risk points are given priority in the inspection plan, followed by medium-risk points, and low-risk points can have their inspection frequency appropriately reduced. On-site safety inspections of each live broadcast point are carried out according to the plan.

[0127] Optionally, in some embodiments of this application, the step of "generating risk levels corresponding to hidden danger information based on historical inspection data and time-series analysis results of the target scenic area, and conducting safety inspections of each live broadcast location based on the risk levels" may specifically include:

[0128] Historical risk information for each live broadcast location was obtained from historical inspection data of the target scenic area.

[0129] Based on historical risk information and time series analysis results, output the fused data corresponding to each live broadcast location;

[0130] Based on the fused data, the risk level corresponding to the hidden danger information is generated;

[0131] Generate inspection strategies corresponding to risk levels, and conduct safety inspections of each live broadcast location based on the inspection strategies.

[0132] Historical inspection data refers to structured data recorded during past safety inspections of the target scenic area, including: the types of hazards that have occurred at each live broadcast point (such as cracks in the boardwalk of a certain academy or congestion at the entrance of a certain park), the time / frequency of the hazard occurrence, the historical risk level (such as a similar crack being classified as a level 3 high risk in 2024), and the handling measures and results (such as a level 3 risk hazard being dealt with within 10 minutes).

[0133] The risk level is a classification of the urgency and severity of the current hidden danger based on historical inspection data and time-series analysis results. Optionally, in some embodiments of this application, it can be divided into the following three levels:

[0134] Level 1 (Low Risk): The potential hazard is dynamically stable, the impact of similar historical cases is minor, and no emergency response is required;

[0135] Level 2 (Medium Risk): The risks are showing signs of worsening, with some historical cases escalating, requiring close monitoring.

[0136] Level 3 (High Risk): The hidden danger is rapidly deteriorating and has caused accidents in historical cases, requiring immediate action.

[0137] Safety inspections refer to targeted on-site inspections and handling activities conducted at each live broadcast location based on the risk level. This includes verifying the status of potential hazards on-site, implementing emergency measures (such as diverting passenger flow and sealing off dangerous areas), and recording the handling results.

[0138] Specifically, historical risk information for each live-streaming location is extracted from historical inspection data. For example, historical inspection data can be filtered by live-streaming location to extract information on similar potential hazards that have occurred at that location. For instance, data on a certain academy's walkway reveals historical risk information: a crack hazard occurred during the 2024 rainy season, with a crack propagation rate of 0.05 cm / minute, which was classified as a Level 3 high risk, ultimately leading to the temporary closure of the walkway. Then, key indicators are extracted from this historical risk information, such as the historical risk threshold for similar cracks at that location. Next, the current dynamic indicators from the time-series analysis results (e.g., the current crack propagation rate of the walkway is 0.04 cm / minute) are compared with the historical risk indicators for that location (e.g., the historical Level 2 risk threshold of 0.03 cm / minute and the Level 3 threshold of 0.05 cm / minute). The similarity between the current hazard trend and historical high-risk, medium-risk, and low-risk cases is assessed. For example, the current crack propagation rate is 0.04 cm / min, falling between the historical level 2 (0.03) and level 3 (0.05) thresholds, and the trend is consistent with the case that upgraded from level 2 to level 3 in 2024 (both cases experienced accelerated propagation during the rainy season). Next, fused data is generated, which includes structured data on historical threshold ranges, current dynamic values, and trend matching. For example, the crack in the Shuyuan boardwalk: currently 0.04 cm / min, located between the historical level 2 and level 3 thresholds, with an 80% match to the high-risk case trend. Then, the fused data is combined with the preset level 3 risk standard.

[0139] If the current dynamic indicator is less than the historical level 1 threshold, and the trend is consistent with historical low-risk cases, it is judged as level 1;

[0140] If the current dynamic indicator is between the historical level one and level two thresholds, or if the trend partially matches the historical medium-risk cases, it is judged as level two;

[0141] If the current dynamic indicator is greater than or equal to the historical level 2 threshold, and the trend highly matches the historical high-risk cases, it is judged as level 3. For example, the fused data of the crack in the Shuyuan Plank Road shows that the current crack rate is 0.04 cm / minute (exceeding the level 2 threshold of 0.03), and the trend matches the high-risk cases with an 80% degree of matching, so it is judged as level 3 high risk.

[0142] Finally, differentiated inspection requirements are matched according to risk level, which can be specifically as follows:

[0143] Level 3 - High Risk: Real-time notification to the security department, who will arrive within 10 minutes, cordon off the site, and carry out repairs.

[0144] Level 2 - Medium Risk: Updates are pushed out every 30 minutes, and inspectors re-inspect every 1-2 hours to record changes in cracks;

[0145] Level 1 - Low Risk: Daily summary and push notifications, routine inspection and confirmation are sufficient the following day.

[0146] Optionally, in some embodiments of this application, the step "generating risk levels corresponding to hazard information based on fused data" may specifically include:

[0147] Determine the degree of risk matching corresponding to the fused data;

[0148] Based on the degree of risk matching and the preset offset range, the risk level corresponding to the hidden danger information is generated.

[0149] The risk matching degree refers to the degree of similarity between the fused data (integrated information of the dynamic characteristics of the current hidden danger and the characteristics of historical risk cases) and historical cases of different risk levels (such as risk level 1, risk level 2, or risk level 3). It is usually expressed as a percentage (such as 80%) or a score (such as 8 points out of 10). Its core is to measure the similarity between the current hidden danger and a certain risk level by comparing the consistency of the key indicators of the current hidden danger (such as the rate of change in passenger flow density, the speed of defect expansion, and the increase in smoke spread), development trends (such as acceleration / deceleration), and environmental factors (such as holidays / weather) with historical cases of each level. For example, if the current smoke spread increase is 5% / minute, and the overlap with the characteristics of historical level 3 high-risk cases (4%-6% / minute, accelerated spread in rainy weather) reaches 90%, then the risk matching degree is 90%.

[0150] Offset range refers to the reasonable range of difference allowed between the fused data and historical risk level characteristics. It is used to cope with fluctuations in real-world scenarios (such as measurement errors and minor environmental changes), and is usually expressed as a percentage or a specific numerical value. For example, in historical level 2 risk, the threshold for guardrail crack propagation speed is 0.03 cm / min, and the preset offset range is ±0.005 cm / min. Then, actual values ​​between 0.025 and 0.035 cm / min are considered to be within the offset range of level 2 risk, and values ​​exceeding this range may be classified as higher or lower levels.

[0151] Specifically, key indicators can be screened from the fused data. Then, historical risk cases of the target live streaming location can be retrieved and categorized into Level 1 (low risk), Level 2 (medium risk), and Level 3 (high risk). Core characteristics of each category can be extracted (e.g., for Level 3 risk cases, the rate of change in passenger flow density is ≥0.5 people / (㎡·minute), peak hours, and accelerating trends). Next, by comparing the core characteristics of the current fused data with the characteristics of historical cases at each level, the similarity is calculated. The higher the percentage of overlap between the current dynamic values, trends, and environmental factors and a certain level of cases, the higher the degree of matching. Finally, the degree of matching between the current fused data and Level 1, Level 2, and Level 3 risks is output (e.g., a matching degree of 20% with Level 1 risk, 30% with Level 2 risk, and 85% with Level 3 risk).

[0152] Furthermore, the preset offset range for each risk level is determined. For example, for Level 3 risk: the current dynamic value is within ±15% of the historical Level 3 threshold; for Level 2 risk: the current dynamic value is within ±20% of the historical Level 2 threshold; and for Level 1 risk: the current dynamic value is within ±30% of the historical Level 1 threshold.

[0153] Next, the risk matching degree is compared with the preset offset range. If the matching degree of a certain level meets the standard and the current dynamic value does not exceed the offset range of that level, it is initially determined to be that level. If the matching degree meets the standard but the dynamic value exceeds the offset range, the level is adjusted according to the direction of the excess (e.g., for level 2 risk, if the dynamic value of 0.6 exceeds the 25% offset range of the level 2 threshold of 0.4, the risk level is adjusted to level 3 risk).

[0154] Optionally, in some embodiments of this application, the step of "generating an inspection strategy corresponding to the risk level" may specifically include:

[0155] Determine the target audience, push frequency, and processing information corresponding to each risk level;

[0156] Based on the target audience, push frequency, and processing information, generate inspection strategies corresponding to the risk level.

[0157] The target audience for the push notifications refers to the responsible entities that receive information on potential risks and their handling instructions. These are typically different functional departments or personnel within the scenic area (such as security departments, dispatch centers, and daily management teams). Their responsibilities are matched to the urgency and difficulty of handling the risk level.

[0158] Push frequency refers to the time interval or trigger condition for sending potential risks information to the target audience. It is set according to the urgency of the risk level (such as real-time push, once every 30 minutes, once a day) to ensure a balance between timeliness and efficiency in information delivery.

[0159] Information processing refers to the specific requirements for handling potential hazards, including response timeliness (e.g., arriving on-site within 10 minutes), operational procedures (e.g., setting up warning signs and evacuating people), and feedback content (e.g., handling results and changes in the hazard), clearly defining "what to do, how to do it, and when to complete it."

[0160] For example, the target audience for the push notification is the scenic area's daily patrol team (responsible for maintaining routine order); the push frequency is a daily summary push (before 18:00 on the same day, the data on visitor gatherings during off-peak hours throughout the day is integrated and pushed); the information to be processed is to focus on checking the area during the next day's patrol (covering within 24 hours), and observe the reasons for the gatherings (such as temporary performances, vendor stalls attracting attention), without intervention but recording is required; if gatherings occur for 3 consecutive days, feedback is given to the operations department to adjust the stall locations;

[0161] The patrol team receives a summary push notification the next day (e.g., "9:00-10:00, 20 people gathered in area A due to food stalls for 30 minutes"); then, the patrol team verifies the situation on-site during their patrol, recording whether the food stalls are in compliance with regulations and whether there is enough space for evacuation; if no abnormalities are found, they can be marked as not requiring action in the system; if gatherings continue, suggestions are pushed to the operations department (e.g., moving the food stalls to an open area).

[0162] The scenic area safety hazard inspection method of this application, after acquiring the live video streams of each live broadcast point in the target scenic area, identifies the hazard information corresponding to each live broadcast point based on a preset recognition model and each live video stream. Then, it performs time-series analysis on the identified hazard information. Finally, based on the historical inspection data of the target scenic area and the time-series analysis results, it generates the risk level corresponding to the hazard information and conducts safety inspections on each live broadcast point according to the risk level. In the scenic area safety hazard inspection scheme based on live broadcasts provided in this application, a preset recognition model is used to identify the hazard information of each live broadcast point, and combined with the historical inspection data of the target scenic area, the risk corresponding to the hazard information is determined. This allows for flexible formulation of inspection strategies, and inspections of different live broadcast points can be carried out based on these strategies. Therefore, this scheme not only improves the efficiency of scenic area safety control but also enhances the reliability of safety control.

[0163] To facilitate better implementation of the scenic area safety hazard inspection method based on live streaming of scenic areas according to the embodiments of this application, this application also provides a scenic area safety hazard inspection device based on live streaming of scenic areas. The meanings of the terms used are the same as in the above-described scenic area safety hazard inspection system based on live streaming of scenic areas, and specific implementation details can be found in the description of the system embodiments.

[0164] Please see Figure 2 , Figure 2 The schematic diagram of the scenic area safety hazard inspection device based on live streaming of scenic areas provided in this application embodiment is shown below. Specifically, the scenic area safety hazard inspection device based on live streaming of scenic areas may include an acquisition module 201, an identification module 202, an analysis module 203, a generation module 204, and an inspection module 205, as follows:

[0165] The acquisition module 201 is used to acquire the live video streams of each live broadcast location in the target scenic area.

[0166] The identification module 202 is used to identify the hidden danger information corresponding to each live broadcast location based on the preset identification model and each live video stream.

[0167] Analysis module 203 is used to perform time-series analysis on the identified hazard information.

[0168] The generation module 204 is used to generate the risk level corresponding to the hidden danger information based on the historical inspection data and time series analysis results of the target scenic area.

[0169] The inspection module 205 is used to conduct safety inspections of each live broadcast location according to the risk level.

[0170] Optionally, in some embodiments of this application, the identification module 202 includes:

[0171] The first acquisition unit is used to acquire a preset recognition model;

[0172] The extraction unit is used to extract video frames from the live video stream of each live broadcast location;

[0173] The output unit is used to input the captured video frames into the recognition model and output the hazard information corresponding to each live broadcast location.

[0174] Optionally, in some embodiments of this application, the interception unit is specifically used for:

[0175] Video frames are extracted from the live video stream of each live location according to the preset frame rate.

[0176] The captured video frames are subjected to image enhancement processing to obtain the processed video frames.

[0177] Optionally, in some embodiments of this application, the analysis module 203 includes:

[0178] The selection unit is used to select consecutive video frames from the live video stream as analysis samples based on the potential hazard information.

[0179] The analysis unit is used to perform time-series analysis on the analysis samples based on the time-series dimension.

[0180] Optionally, in some embodiments of this application, the analysis unit is specifically used for:

[0181] Calculate the rate of change of passenger flow density for the analyzed sample per unit time, and output the passenger flow risk corresponding to the analyzed sample based on the rate of change of passenger flow density, and / or;

[0182] Monitor the defect changes of the first target object in the analysis sample within a unit of time, and output the defect risk of the object corresponding to the analysis sample based on the defect changes of the target object, and / or;

[0183] The system monitors and analyzes the pixel percentage of the second target object in each image frame and outputs the object spread risk corresponding to the analysis sample based on the increase of the second target object within a unit of time.

[0184] Optionally, in some embodiments of this application, the inspection module 205 includes:

[0185] The second acquisition unit is used to obtain historical risk information corresponding to each live broadcast point from the historical inspection data of the target scenic area.

[0186] The output unit is used to output the fused data corresponding to each live broadcast location based on historical risk information and time series analysis results.

[0187] The generation unit is used to generate the risk level corresponding to the hidden danger information based on the fused data;

[0188] The inspection unit is used to generate inspection strategies corresponding to risk levels and to conduct safety inspections of each live broadcast location based on the inspection strategies.

[0189] Optionally, in some embodiments of this application, the generation unit may specifically be used for:

[0190] Determine the degree of risk matching corresponding to the fused data;

[0191] Based on the degree of risk matching and the preset offset range, the risk level corresponding to the hidden danger information is generated.

[0192] Optionally, in some embodiments of this application, the inspection unit may specifically be used for:

[0193] Determine the target audience, push frequency, and processing information corresponding to each risk level;

[0194] Based on the target audience, push frequency, and processing information, generate inspection strategies corresponding to the risk level.

[0195] This application provides a scenic area safety hazard inspection device. After the acquisition module 201 acquires the live video streams from each live streaming point in the target scenic area, the identification module 202 identifies the hazard information corresponding to each live streaming point based on a preset identification model and the live video streams. Next, the analysis module 203 performs time-series analysis on the identified hazard information. Finally, the generation module 204 generates the risk level corresponding to the hazard information based on the historical inspection data of the target scenic area and the time-series analysis results. The inspection module 205 then conducts safety inspections of each live streaming point according to the risk level. In the scenic area safety hazard inspection scheme based on live streaming provided in this application, a preset identification model is used to identify the hazard information at each live streaming point, and combined with the historical inspection data of the target scenic area, the risk corresponding to the hazard information is determined. This allows for flexible formulation of inspection strategies, and inspections can be conducted on different live streaming points based on these strategies. Therefore, this scheme not only improves the efficiency of scenic area safety control but also enhances the reliability of safety control.

[0196] Furthermore, embodiments of this application also provide an electronic device, such as... Figure 3As shown, it illustrates a structural schematic diagram of the electronic device involved in the embodiments of this application, specifically:

[0197] The electronic device may include components such as a processor 301 with one or more processing cores, a memory 302 with one or more processor-readable storage media, a power supply 303, and an input unit 304. Those skilled in the art will understand that... Figure 3 The electronic device structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements. Wherein:

[0198] The processor 301 is the control center of the electronic device. It connects various parts of the electronic device via various interfaces and lines. By running or executing software programs and / or modules stored in the memory 302, and by calling data stored in the memory 302, it performs various functions and processes data, thereby providing overall monitoring of the electronic device. Optionally, the processor 301 may include one or more processing cores; preferably, the processor 301 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may not be integrated into the processor 301.

[0199] The memory 302 can be used to store software programs and modules. The processor 301 executes various functional applications and a scenic area safety hazard inspection method based on live streaming by running the software programs and modules stored in the memory 302. The memory 302 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the electronic device, etc. In addition, the memory 302 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory 302 may also include a memory controller to provide the processor 301 with access to the memory 302.

[0200] The electronic device also includes a power supply 303 that supplies power to various components. Preferably, the power supply 303 can be logically connected to the processor 301 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 303 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0201] The electronic device may also include an input unit 304, which can be used to receive input digital or character information and generate keyboard, mouse, joystick, optical or trackball signal inputs related to user settings and function control.

[0202] Although not shown, the electronic device may also include a display unit, etc., which will not be described in detail here. Specifically, in this embodiment, the processor 301 in the electronic device loads the executable files corresponding to the processes of one or more applications into the memory 302 according to the following instructions, and the processor 301 runs the applications stored in the memory 302 to realize various functions, as follows:

[0203] Acquire live video streams from each live streaming location in the target scenic area; based on a pre-set recognition model and each live video stream, identify the hazard information corresponding to each live streaming location; perform time-series analysis on the identified hazard information; generate the risk level corresponding to the hazard information based on the historical inspection data of the target scenic area and the time-series analysis results, and conduct safety inspections on each live streaming location according to the risk level.

[0204] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0205] This application embodiment, after acquiring the live video streams from each live streaming location of the target scenic area, identifies the corresponding hidden danger information for each live streaming location based on a preset recognition model and the live video streams. Then, it performs time-series analysis on the identified hidden danger information. Finally, based on the historical inspection data of the target scenic area and the time-series analysis results, it generates the risk level corresponding to the hidden danger information and conducts safety inspections on each live streaming location according to the risk level. In the scenic area safety hidden danger inspection scheme based on live streaming provided in this application, a preset recognition model is used to identify the hidden danger information at each live streaming location, and combined with the historical inspection data of the target scenic area, the risk corresponding to the hidden danger information is determined. This allows for flexible formulation of inspection strategies, and inspections can be conducted on different live streaming locations based on these strategies. Therefore, this scheme not only improves the efficiency of safety control in scenic areas but also enhances the reliability of safety control.

[0206] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be performed by instructions, or by instructions controlling related hardware. These instructions can be stored in a processor-readable storage medium and loaded and executed by a processor.

[0207] Therefore, embodiments of this application provide a storage medium storing multiple instructions that can be loaded by a processor to execute steps in any of the scenic area safety hazard inspection methods based on live streaming provided in this application. For example, the instructions can execute the following steps:

[0208] Acquire live video streams from each live streaming location in the target scenic area; based on a pre-set recognition model and each live video stream, identify the hazard information corresponding to each live streaming location; perform time-series analysis on the identified hazard information; generate the risk level corresponding to the hazard information based on the historical inspection data of the target scenic area and the time-series analysis results, and conduct safety inspections on each live streaming location according to the risk level.

[0209] For details on the implementation of each of the above operations, please refer to the previous examples, which will not be repeated here.

[0210] The storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0211] Since the instructions stored in the storage medium can execute the steps in any of the scenic area safety hazard inspection methods based on live streaming provided in the embodiments of this application, the beneficial effects that any of the scenic area safety hazard inspection methods based on live streaming provided in the embodiments of this application can achieve can be realized. For details, please refer to the previous embodiments, which will not be repeated here.

[0212] The foregoing has provided a detailed description of a method, device, electronic device, and storage medium for inspecting safety hazards in scenic areas based on live streaming of scenic areas, as provided in the embodiments of this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for inspecting safety hazards in scenic areas, characterized in that, include: Obtain live video streams from various live streaming locations within the target scenic area; Based on the preset recognition model and each of the live video streams, identify the hidden danger information corresponding to each live broadcast location; Perform time-series analysis on the identified hazard information; Based on the historical inspection data and time-series analysis results of the target scenic area, a risk level corresponding to the hidden danger information is generated, and safety inspections are carried out on each live broadcast location according to the risk level.

2. The method for inspecting safety hazards in scenic areas according to claim 1, characterized in that, The process of identifying potential hazard information corresponding to each live streaming location based on a preset identification model and each of the live video streams includes: Obtain the preset recognition model; Extract video frames from the live video streams of each live streaming location; The captured video frames are input into the recognition model, which outputs the hazard information corresponding to each live broadcast location.

3. The method for inspecting safety hazards in scenic areas according to claim 2, characterized in that, The step of extracting video frames from the live video streams at each live streaming location includes: Video frames are extracted from the live video stream of each live location according to the preset frame rate. The captured video frames are subjected to image enhancement processing to obtain the processed video frames.

4. The method for inspecting safety hazards in scenic areas according to claim 1, characterized in that, The time-series analysis of the identified hazard information includes: Based on the potential hazard information, consecutive video frames are selected from the live video stream as analysis samples; The analysis samples are subjected to time-series analysis based on the time-series dimension.

5. The method for inspecting safety hazards in scenic areas according to claim 4, characterized in that, The time-series analysis of the analysis samples based on the time-series dimension includes: Calculate the rate of change of passenger flow density of the analyzed sample per unit time, and output the passenger flow risk corresponding to the analyzed sample based on the rate of change of passenger flow density, and / or; Monitor the defect changes of the first target object in the analysis sample within a unit time period, and output the object defect risk corresponding to the analysis sample based on the defect changes of the target object, and / or; The pixel percentage of the second target object in each image frame of the analysis sample is monitored, and the object spread risk corresponding to the analysis sample is output according to the increase of the second target object in a unit time.

6. The method for inspecting safety hazards in scenic areas according to claim 1, characterized in that, The process involves generating a risk level corresponding to the hidden danger information based on historical inspection data and time-series analysis results of the target scenic area, and conducting safety inspections of each live broadcast location based on the risk level, including: Historical risk information corresponding to each live broadcast location is obtained from the historical inspection data of the target scenic area. Based on the historical risk information and time series analysis results, output the fused data corresponding to each live broadcast location; Based on the fused data, a risk level corresponding to the hidden danger information is generated; Generate inspection strategies corresponding to the risk levels, and conduct security inspections on each live broadcast location based on the inspection strategies.

7. The method for inspecting safety hazards in scenic areas according to claim 6, characterized in that, The step of generating the risk level corresponding to the hidden danger information based on the fused data includes: Determine the degree of risk matching corresponding to the fused data; Based on the risk matching degree and the preset offset range, the risk level corresponding to the hidden danger information is generated.

8. The method for inspecting safety hazards in scenic areas according to claim 6, characterized in that, The process of generating the inspection strategy corresponding to the risk level includes: Determine the target audience, push frequency, and processing information corresponding to each risk level; Based on the push target, push frequency, and processing information, an inspection strategy corresponding to the risk level is generated.

9. A scenic area safety hazard inspection device based on live streaming, characterized in that, include: The acquisition module is used to acquire live video streams from various live streaming locations within the target scenic area. The identification module is used to identify potential hazard information corresponding to each live broadcast location based on a preset identification model and each of the live video streams. The analysis module is used to perform time-series analysis on the identified hazard information; The generation module is used to generate the risk level corresponding to the hidden danger information based on the historical inspection data and time series analysis results of the target scenic area. The inspection module is used to conduct safety inspections of each live broadcast location based on the risk level.

10. An electronic device, characterized in that, include: A memory, a processor, and a processor program stored in the memory and executable on the processor, wherein the processor executes the program as described in any one of claims 1 to 8, the steps of the scenic area safety hazard inspection method based on live streaming of scenic areas.

11. A storage medium, characterized in that, The system stores a computer processing program capable of being loaded by a processor and executing the scenic area safety hazard inspection method based on live streaming as described in any one of claims 1 to 8.

Images