Intelligent forest fire prevention propaganda system based on video monitoring

Abstract

This invention relates to the field of forest fire prevention technology and discloses an intelligent forest fire prevention publicity system based on video surveillance. The system includes: a front-end sensing and acquisition module deployed at monitoring points to collect video streams and environmental data; a behavior feature analysis module to identify pedestrian / vehicle targets and generate their behavior feature tags; a comprehensive risk assessment and decision-making module to fuse and analyze behavior feature tags and environmental data, identify spatiotemporal coupled risks of environment and behavior, generate a comprehensive risk level and matching warning response instructions; a multimodal fusion publicity module to play voice and dynamic graphic warning signals; and a remote interaction and control module for three-dimensional visualization and supporting two-way communication between pedestrians and administrators. This invention achieves a transformation from passive publicity to proactive on-demand response, and from single risk perception to coupled risk quantitative assessment, significantly improving the accuracy, intelligence level, and emergency interaction capabilities of forest fire prevention publicity.

Description

Technical Field

[0001] This invention relates to the field of forest fire prevention technology, and in particular to an intelligent forest fire prevention publicity system based on video surveillance. Background Technology

[0002] Currently, the main methods of forest fire prevention publicity are static warning signs, fixed voice poles, or manual patrols and announcements. These traditional methods lack the ability to perceive the situation on the ground in real time, resulting in wasted electricity and resources during periods when no one is around. Furthermore, they cannot provide timely and targeted warnings and interventions when people exhibit abnormal behavior, making the publicity effect disproportionate to the resource investment.

[0003] In the field of forest fire monitoring technology, existing solutions fall into two categories: one is to deploy environmental sensors to monitor fire risk factors such as temperature, humidity, and wind force; the other is to use video surveillance for simple motion detection or smoke and fire identification. However, the vast majority of forest fires are the result of the combined effects of human activities and harsh environments. Existing technologies fail to perform spatiotemporal correlation analysis between the dynamic behavioral characteristics of on-site personnel and environmental data, and the risk assessment models are too simplistic, making it difficult to output accurate dynamic risk levels.

[0004] Currently available intelligent advertising equipment, although integrating voice broadcasting and display screens, still operates in a fixed mode. For example, the volume of the voice broadcast and the brightness of the screen display remain constant, and voice and text usually operate independently or in simple synchronization, lacking the ability to coordinate and adjust the output mode according to different risk levels.

[0005] For front-end propaganda equipment scattered across vast forest areas, managers often lack efficient and intuitive remote control methods. They can usually only view historical video recordings or equipment status. When violations are detected, there is a lack of convenient remote communication channels to promptly dissuade them. Furthermore, when tourists encounter sudden dangers, they cannot find a way to establish direct contact with management personnel. Summary of the Invention

[0006] This invention provides an intelligent forest fire prevention publicity system based on video surveillance to solve existing technical problems.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides an intelligent forest fire prevention publicity system based on video surveillance, comprising: The front-end sensing and acquisition module is deployed at forest fire monitoring points to collect video stream data and environmental data of the monitoring area in real time. The behavior feature analysis module is used to identify pedestrian and vehicle targets in the video stream data, analyze their current location and intention orientation, and assign behavior feature labels based on the intention orientation; The integrated risk assessment and decision-making module is used to integrate and analyze the behavioral feature tags and the environmental data to generate an integrated risk level, and to match the corresponding warning theme, warning intensity and output sequence from the preset publicity strategy library according to the integrated risk level to generate a warning response instruction; The multimodal fusion publicity module is used to respond to the warning response command and play voice warning signals and dynamic graphic warning signals at the forest fire prevention monitoring point; The remote interaction and control module is used to receive and display the video stream data, the behavioral feature tags and the comprehensive risk level, and send the publicity content update instructions and remote broadcast instructions to the multimodal fusion publicity module. It also supports two-way communication between pedestrians and managers.

[0008] The beneficial effects of the technical solution provided by this invention include at least the following: This invention accurately identifies pedestrians and vehicles from video streams and analyzes their dwell time, movement trajectory, and other characteristics to determine their intentions. Instead of indiscriminately advertising to all entrants, the system triggers differentiated warning strategies based on their behavioral characteristic tags. This hierarchical triggering mechanism based on behavioral intention significantly improves the targeting and timeliness of interventions for high-risk behaviors while saving energy and reducing consumption.

[0009] This invention creatively constructs a spatiotemporally coupled risk quantification model of environment and behavior. This module can not only align environmental and behavioral data, but also identify the correlation between the two through a coupled risk discrimination model to form coupled risk event pairs. At the same time, dynamic weight adjustment is introduced, which enables the system to dynamically adjust the weights of environmental anomalies and behavioral anomalies in risk fusion calculation based on the strength of their statistical correlation, so as to generate a more realistic and dynamically changing comprehensive risk level, which greatly improves the scientificity and accuracy of fire risk assessment.

[0010] This invention, through modal collaborative control and time-series synchronous control, can execute complex audiovisual propaganda strategies based on the comprehensive risk level, enhancing the impact of warning information from multiple sensory dimensions. At the same time, combined with the environmental perception submodule and the output parameter dynamic adjuster, the system can automatically optimize the brightness of the display screen and the volume of the voice broadcast according to the ambient light intensity and background noise, ensuring that the propaganda content can always reach the target audience with the best effect, reflecting intelligent human-centered care.

[0011] This invention provides managers with a comprehensive management platform integrating three-dimensional situational awareness, remote command, and emergency interaction by constructing a remote interaction and control module. It breaks the limitation of the one-way output of traditional publicity systems, provides the public with an emergency help channel, and, combined with the function of automatically popping up on-site video, realizes a closed loop of emergency response from help request to handling. It upgrades fire prevention publicity nodes into comprehensive intelligent nodes integrating publicity, management, service, and emergency response. Attached Figure Description

[0012] To more clearly illustrate the technical solutions in the embodiments of the present invention, 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 the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a system structure diagram provided in an embodiment of the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in further detail below with reference to the accompanying drawings.

[0015] This embodiment provides an intelligent forest fire prevention publicity system based on video surveillance. Please refer to... Figure 1 This is a system structure diagram provided in an embodiment of the present invention.

[0016] An intelligent forest fire prevention publicity system based on video surveillance, as described in this embodiment, includes: 1. Front-end sensing and acquisition module, deployed at forest fire monitoring points, used to collect video stream data and environmental data of the monitoring area in real time; The behavioral feature analysis module is used to identify pedestrian and vehicle targets in video stream data, analyze their current location and intent orientation, and assign behavioral feature labels based on intent orientation. The comprehensive risk assessment and decision-making module is used to integrate and analyze behavioral characteristic tags and environmental data to generate a comprehensive risk level. Based on the comprehensive risk level, it matches the corresponding warning theme, warning intensity and output sequence from the preset publicity strategy library to generate warning response instructions. The multimodal integrated publicity module is used to respond to warning commands and play voice warning signals and dynamic graphic warning signals at forest fire monitoring points; The remote interaction and control module is used to receive and display video stream data, behavioral feature tags and comprehensive risk levels, and send instructions to update publicity content and remote broadcast instructions to the multimodal fusion publicity module. It also supports two-way communication between pedestrians and managers.

[0017] In one embodiment of the present invention, the front-end sensing and acquisition module includes the following sub-modules: The video acquisition submodule is deployed on the top of the pole at the forest fire monitoring point and is used to collect video stream data of the monitoring area in real time. The environmental monitoring submodule, integrated with the video acquisition submodule and installed on the same pole, is used to collect environmental data of the monitoring area, including temperature, humidity and wind speed. The infrared sensing submodule, deployed in conjunction with the video acquisition submodule, is used to detect whether pedestrians or vehicles are passing through the monitored area. The edge computing submodule, connected to the video acquisition submodule and the infrared sensing submodule, is used for front-end preprocessing of video stream data, including motion detection and human / vehicle recognition, and triggers video stream data storage and uploading only when a pedestrian or vehicle is detected.

[0018] It should be noted that in this embodiment, the video acquisition submodule uses a network camera with a 4-megapixel (2560×1440 resolution) CMOS image sensor. The lens focal length can be selected as 4mm / 6mm / 8mm, and the field of view is configured from 60° to 120° according to the width of the monitoring area. It supports H.265 / H.264 video encoding formats and has an infrared night vision function of 10~30 meters. This submodule is deployed on the top of the pole and fixed by a universal bracket. The pitch angle and horizontal rotation angle can be adjusted according to the monitoring area. The installation height is usually 3-5 meters to ensure that the field of view covers the entrance and key paths of the monitoring area. It outputs RTSP video stream in real time and transmits it to the edge computing submodule via network cable.

[0019] In this embodiment, the environmental monitoring submodule uses an integrated environmental sensor, which integrates a digital temperature sensor, a capacitive humidity sensor, and a three-cup anemometer or ultrasonic anemometer. This submodule is integrated with the video acquisition submodule on the same pole, maintaining an appropriate distance from the camera (usually ≥0.5 meters) to avoid the camera obstructing the accuracy of wind field measurement. The installation height is not less than 2 meters, and data is collected once per minute. The data is transmitted to the edge computing submodule via RS485 or Modbus protocol. Those skilled in the art can select commercially available mature environmental sensor products according to actual monitoring needs and install and debug them according to the product manual.

[0020] The infrared sensing submodule uses a passive infrared (PIR) sensor, which is deployed on a pole in conjunction with the video acquisition submodule. The installation height is approximately 1.5 to 2 meters, tilted downwards at 5° to 10° to ensure coverage of the monitoring area entrance and pedestrian / vehicle traffic paths. Installation should avoid direct sunlight. When a pedestrian or vehicle enters the detection area, the infrared radiation emitted by the person / vehicle is focused by the Fresnel lens built into the sensor, causing a pyroelectric effect. The sensor outputs a high-level signal, which is amplified and compared before outputting a digital trigger signal. The PIR sensor's detection distance adjustment, angle coverage design, and anti-interference measures are all common knowledge in sensor applications. In actual deployment, the sensitivity potentiometer can be adjusted according to the site environment to set a suitable detection distance.

[0021] The edge computing submodule runs a lightweight embedded Linux system and deploys the following software functions: RTSP video stream reception, motion detection (frame difference or background subtraction method, used to detect whether there are moving targets in the picture as an auxiliary trigger condition), human / vehicle recognition (lightweight deep learning model, used to identify whether there are pedestrians or vehicles in the picture, reducing false triggers caused by animals, tree branches swaying, etc.), image capture, and video recording. The above edge computing hardware selection, software framework, and trigger logic design are all conventional technologies in the field of embedded system development. Those skilled in the art can select commercially available edge computing modules according to actual performance and cost requirements, and perform secondary development according to the SDK provided by the chip manufacturer to realize the functions. The motion detection and human recognition algorithms can adopt mature models in open source algorithm libraries.

[0022] To more clearly illustrate the data coupling relationship between the sub-modules, the following describes the complete workflow using a typical application scenario as an example: Scene: A pedestrian enters a forest fire monitoring area. When a pedestrian enters the detection area of ​​the infrared sensing submodule, the PIR sensor outputs a high-level signal to the GPIO input port of the edge computing submodule; The edge computing submodule is woken up by a GPIO interrupt, switches from low-power standby mode to working mode, and starts receiving camera video streams; The video acquisition submodule starts transmitting real-time video streams, while the edge computing submodule decodes video frames and runs motion detection and human recognition algorithms simultaneously. Motion detection identified a moving target in the image, and the human recognition model identified the target, confirming the presence of a pedestrian. The edge computing submodule retrieves the cached video of the 30 seconds before the trigger (pre-recorded via a circular buffer) and starts recording the video of the 30 seconds after the trigger, capturing 3 JPEG images at 0.5-second intervals. Environmental data association: The edge computing submodule reads the current temperature, humidity, and wind speed data from the environmental monitoring submodule, as well as the historical environmental data from the 5 minutes prior to the trigger, and packages it with the video data; The edge computing submodule uploads data packets to the remote management platform via the 4G module. If the network is unavailable, the packaged data packets are marked as pending upload and will automatically resume uploading once the network is restored. After the upload is complete, the edge computing submodule turns off the camera power and enters a low-power standby mode, waiting for the next infrared trigger.

[0023] In one embodiment of the present invention, the behavior feature analysis module is connected to the front-end perception and acquisition module, and includes the following sub-modules: The target detection and tracking submodule is connected to the front-end perception and acquisition module. It is used to identify pedestrian and vehicle targets from video stream data, track their real-time position, direction of movement and speed, and generate motion trajectory data containing the target ID. The dwell time analysis submodule is used to calculate the dwell time of pedestrian and vehicle targets within the monitoring area; The trajectory pattern recognition submodule is used to analyze the target's motion trajectory pattern, including path direction, number of hovering times, and entry / exit direction; The intent orientation determination submodule is used to fuse dwell time with motion trajectory pattern features to generate corresponding behavioral feature labels.

[0024] It should be noted that in this embodiment, the target detection and tracking submodule is used to detect pedestrian and vehicle targets in real time from video stream data, and continuously track the detected targets to generate trajectory data containing target ID, real-time position, direction of movement, and speed of movement. The target detection algorithm is implemented using a lightweight YOLO series target detection model, and the multi-target tracking algorithm is implemented using ByteTrack or DeepSORT algorithms. The YOLO series target detection algorithm and the ByteTrack / DeepSORT multi-target tracking algorithm are mature technologies in the field of computer vision. Those skilled in the art can obtain the algorithm source code and pre-trained models through open source communities such as GitHub, and adapt and optimize them according to the specific hardware platform to achieve the required functions.

[0025] The dwell time analysis submodule is used to calculate the dwell time of each target within the monitoring area and classify the dwell time into different levels according to preset time thresholds. Specific implementation methods include: (1) Calculation of stay duration: When the target first appears in the monitoring area (trajectory creation), record the current timestamp as enter_time; When the target disappears from the monitoring area (the trajectory ends), record the current timestamp as leave_time; Calculate the stay duration: stay_duration = leave_time - enter_time (unit: seconds); For targets still within the monitoring area, their real-time stay duration is calculated by subtracting the entry time from the current time.

[0026] (2) Dwell Time Classification: A threshold for the preset dwell time range is used to map the dwell time to the corresponding level, as shown in Table 1: Table 1: Mapping of Stay Duration Levels The thresholds for the aforementioned dwell time range can be dynamically adjusted according to different monitoring areas. For example, the core fire protection zone can be set with stricter thresholds (such as 1 minute being considered an abnormal stay), while the thresholds for general buffer zones can be relaxed. The thresholds can be updated through the remote management and control platform.

[0027] The trajectory pattern recognition submodule is used to analyze the target's movement trajectory pattern, identify features such as path direction, number of hovering times, and entry / exit direction, and compare them with a preset risk behavior trajectory database. The specific implementation method is shown in Table 2: Table 2: Schematic Table of Trajectory Feature Extraction Methods Those skilled in the art should know that trajectory feature extraction can be achieved using geometric calculation methods, and wandering detection and direction reversal detection can be achieved by analyzing the directional changes of trajectory points. The above algorithms are all common knowledge in the field and can be implemented using open source libraries such as OpenCV and Shapely.

[0028] The intent orientation determination submodule is used to fuse the dwell time analysis results with the trajectory pattern recognition results to generate the target's intent orientation label, which is output as a behavioral feature label. A decision tree model or rule engine is used to fuse and determine multi-dimensional features. An example of a rule for establishing an intent orientation label system is given in Table 3: Table 3: Schematic diagram of the rules for establishing an intention orientation labeling system The rule engine can be implemented using open-source rule engines such as Drools and EasyRules, or it can be implemented using simple if-else logic. The decision tree model can be trained using machine learning libraries such as Scikit-learn. All of the above technologies are mature technologies in this field, and those skilled in the art can choose the appropriate implementation method according to actual needs. In addition, the identification of fire-using behaviors such as smoking needs to integrate a dedicated fire-using behavior identification model into the behavior feature analysis module. This model can use deep learning-based image classification or object detection algorithms. By training an image dataset containing fire-using behaviors such as smoking, lighters, and matches, real-time identification of fire-using behaviors in video frames can be achieved. Those skilled in the art can train a recognition model that meets the accuracy requirements using publicly available fire source detection datasets.

[0029] To more clearly illustrate the data coupling relationship between the sub-modules, the following describes the complete workflow using a typical application scenario as an example: Scenario: A pedestrian lingers within the monitored area for more than 5 minutes; The video acquisition module inputs the video stream into the target detection and tracking submodule. The target detection model identifies the pedestrian target and assigns the target ID "P001". The tracking algorithm continuously updates the target position and generates a sequence of trajectory points. The dwell time analysis submodule records the target entry time T0 and updates the target dwell time every frame. When the dwell time reaches 2 minutes, the "level_2" flag is triggered, and when the dwell time reaches 5 minutes, it is upgraded to the "level_3" flag. The trajectory pattern recognition submodule analyzes the target trajectory in real time. When it detects that the number of direction reversals reaches 4, it is judged as "wandering behavior". The intent orientation determination submodule receives the dwell time level (level_3) and abnormal pattern (loitering behavior), matches it according to the establishment rules of the intent orientation label system: "long-term stay + loitering behavior → potential risk behavior", generates the intent label "potential_risk", and outputs the encapsulated behavioral feature label to the comprehensive risk assessment module.

[0030] In one embodiment of the present invention, the integrated risk assessment and decision-making module is connected to the front-end perception and acquisition module and the behavioral feature analysis module, respectively, and includes a spatiotemporally coupled risk identification submodule, which is used for: The received environmental data and behavioral feature labels are aligned with timestamps and spatial coordinates to generate a spatiotemporally synchronized environment-behavior joint dataset. Environmental anomaly features and behavioral anomaly features are extracted from the joint environment-behavior dataset. Environmental anomaly features include at least one of temperature anomaly, humidity anomaly, and wind speed anomaly. Behavioral anomaly features include at least one of abnormal loitering, wandering trajectory, and fire use behavior including smoking. By using a built-in coupling risk discrimination model, the correlation between environmental anomalies and behavioral anomalies that occur simultaneously within the same spatiotemporal window is identified, and it is determined whether the two constitute a coupling risk. The environmental anomalies and behavioral anomalies that are determined to constitute a coupling risk are encapsulated as coupling risk event pairs. The coupling risk event pairs are used to input a risk fusion calculation model to generate a comprehensive risk level.

[0031] It should be noted that in this embodiment, the spatiotemporal coupling risk identification submodule first maintains an environmental data cache for the most recent 24 hours, which is stored according to the monitoring point ID and time index. When a new behavioral feature tag is received, the timestamp and location coordinates of the behavior are first extracted. Then, the existing environmental data within 60 seconds before and after the time of the behavior is searched in the environmental data cache. If the time difference of the found environmental data is within the allowable range, it is considered that the time match is successful.

[0032] Spatial matching uses a distance threshold method. The system pre-stores the geographic coordinates of each monitoring point, calculates the Euclidean distance between the location where the behavior occurred and each monitoring point, finds the nearest monitoring point, and if the distance is less than the preset spatial threshold (default 50 meters), it is considered a successful spatial match, and the environmental data of the monitoring point is associated with the behavior data.

[0033] Behavioral feature tags that are successfully matched in both time and space are encapsulated with environmental data into environmental-behavioral joint data, which includes information such as behavioral data, environmental data, time difference, and spatial distance. If time matching or spatial matching fails, the behavioral data will not be used for coupling risk assessment. Those skilled in the art should understand that the specific values ​​of the time window and the preset spatial threshold can be adjusted according to the actual situation of the monitoring area. For example, the spatial threshold can be appropriately relaxed in open areas of forest, and the time window can be tightened in key fire prevention areas.

[0034] In this embodiment, the system has a preset environmental anomaly judgment threshold table, which can be dynamically adjusted according to the season, region, and fire risk level issued by the environmental department. A typical environmental anomaly threshold setting is as follows: Temperature anomaly determination: When the monitored temperature is ≥38℃, it is determined to be a high temperature anomaly; when it is <-15℃, it is determined to be a low temperature anomaly. Humidity anomaly determination: When the monitored relative humidity is <20%, it is determined to be dry anomaly; when it is ≥90%, it is determined to be high humidity anomaly. Wind speed anomaly judgment: When the monitored wind speed is ≥12 m / s (approximately level 6 wind), it is judged as a strong wind anomaly; For each environmental monitoring data point, the system compares it one by one with the above-mentioned environmental anomaly thresholds, marks environmental elements that exceed the environmental anomaly threshold range as environmental anomaly features, and records the anomaly type, actual temperature / humidity / wind speed values, and environmental anomaly threshold.

[0035] Abnormal behavioral features are directly extracted from the behavioral feature labels output by the behavioral feature analysis module. The system focuses on the following types of abnormal behavior: Abnormal loitering features: When the dwell time level in the behavioral feature label is level_2 (dwell time 2-5 minutes) or level_3 (dwell time more than 5 minutes), it is extracted as an abnormal loitering feature; Loitering trajectory features: When the abnormal pattern in the behavioral feature label contains loitering behavior, it is extracted as a loitering trajectory feature, and the number of loitering is recorded; Fire use behavior characteristics: When the abnormal pattern in the behavior characteristic label includes smoking behavior or when smoking, open flame and other fire use behaviors are identified by a special fire source detection model, they are extracted as fire use behavior characteristics.

[0036] In this embodiment, the built-in coupling risk discrimination model identifies the correlation between environmental anomalies and behavioral anomalies that occur simultaneously within the same spatiotemporal window, and determines whether the two constitute a coupling risk. The system defines the following spatiotemporal window as the basis for coupling determination: Time window: 30 minutes, meaning that if the time difference between the occurrence of environmental abnormalities and behavioral abnormalities is within 30 minutes, they are considered to occur simultaneously in time; Spatial window: 50 meters, meaning that if the distance between the environmental anomaly monitoring point and the location where the behavior occurred is within 50 meters, it is considered to occur simultaneously in space; When environmental anomalies and behavioral anomalies occur simultaneously within the same spatiotemporal window, they are included in the coupling risk assessment. The system uses a rule engine to assess coupling risk, and the following typical coupling risk assessment rules are preset: Coupled rule of high temperature and drought with smoking behavior: When environmental anomalies include high temperature anomalies or dryness anomalies, and behavioral anomalies include smoking behavior, it is judged as a high-risk coupled event. This rule is based on common sense knowledge of forest fire prevention, that is, smoking behavior is very likely to cause forest fires under high temperature and dry weather. Strong wind and abnormal loitering coupling rule: When environmental anomalies include strong wind anomalies, and behavioral anomalies include abnormal loitering, it is judged as a medium-risk coupling event. People staying for a long time in strong wind weather poses safety hazards, such as being injured by falling tree branches. Dryness and loitering behavior coupling rule: When environmental anomalies include dryness anomalies and behavioral anomalies include loitering behavior, it is judged as a medium-risk coupling event. People or vehicles loitering for a long time in a dry environment may increase the risk of unintentionally causing a fire. High Temperature and Abnormal Stay Coupling Rule: When an abnormal environment includes an abnormal high temperature, and an abnormal behavior includes abnormal stay, it is judged as a medium-risk coupling event. People staying in high temperature weather for a long time pose a health risk such as heatstroke.

[0037] The above rules can be added, deleted, modified, and queried according to the actual application scenario. The rule engine can be managed in the form of decision tables or rule files, which makes it easy to dynamically adjust the rules without modifying the code.

[0038] In one embodiment of the present invention, the comprehensive risk assessment and decision-making module further includes a dynamic weight adjustment submodule, which is used for: The frequency of occurrence of coupled risk events is statistically analyzed within a continuous spatiotemporal window, and the correlation coefficient between environmental anomalies and behavioral anomalies is calculated. Based on the built-in dynamic weight function library, the dynamic weight coefficients of environmental factors and behavioral factors within the current spatiotemporal window are calculated according to the frequency of occurrence and the correlation coefficient. When the frequency of occurrence exceeds the preset frequency threshold and the correlation coefficient exceeds the preset correlation threshold, the dynamic weight coefficient of the behavioral factor is automatically increased according to the preset increment function. The dynamic weighting coefficients are output to the risk fusion calculation model, which is used to weight and fuse environmental anomaly features and behavioral anomaly features when generating a comprehensive risk level.

[0039] It should be noted that in this embodiment, the system defines a continuous spatiotemporal window as the basic unit of statistical analysis. The length of the time window can be set according to the actual situation of the monitoring area. Unlike the spatiotemporal window used for coupling risk definition mentioned above, the time window in this embodiment is set to 24 hours by default, that is, from 0:00 a.m. to 0:00 a.m. the next day is a statistical cycle. The spatial window is based on the monitoring area of ​​a single monitoring point, that is, each monitoring point is statistically analyzed independently. Similarly, in key fire prevention areas, the time window can be dynamically shortened to 12 hours or 6 hours to improve response sensitivity. In low-risk fire prevention areas, the time window can be dynamically extended to 48 hours or 72 hours to smooth short-term fluctuations. Within each spatiotemporal window, the system counts the frequency of the following three types of events: Frequency of environmental anomalies: refers to the total number of various environmental anomalies (high temperature, dryness, strong wind, etc.) that occur at this monitoring point; Frequency of abnormal behavior events: refers to the total number of various abnormal behaviors (abnormal loitering, wandering, fire use, etc.) that occur at the monitoring point; Coupling risk event frequency: refers to the total number of coupled risk event pairs that occur at this monitoring point. The same behavior may trigger multiple coupling rules at the same time, but it is only counted as one coupled risk event pair.

[0040] The system uses the Pearson correlation coefficient to calculate the linear correlation between environmental anomalies and behavioral anomalies. The calculation steps are as follows: (1) Further subdivide the time window into time segments (e.g., 1 hour), count the number of environmental abnormal events and behavioral abnormal events in each segment, and form two time series data; (2) Calculate the covariance and standard deviation of the two time series, and then obtain the Pearson correlation coefficient. Its value ranges from -1 to 1. A positive value indicates a positive correlation, a negative value indicates a negative correlation, and the larger the absolute value, the stronger the correlation. For forest fire prevention scenarios, focus on positive correlation, that is, when environmental anomalies increase, behavioral anomalies also increase simultaneously, such as increased human activity during the afternoon high temperature period, or increased smoking behavior in dry weather, etc. (3) After each statistical period ends, the correlation coefficients of the past N (default N=7) time windows are recalculated; Those skilled in the art should understand that the Pearson correlation coefficient is a well-known calculation method in statistics and can be calculated directly using tools such as Excel, Python's NumPy library, or the R language.

[0041] In this embodiment, the system presets the following two thresholds as triggering conditions for weight adjustment: Frequency threshold: When the frequency of occurrence of a coupled risk event within a single spatiotemporal window exceeds a preset frequency threshold (default is 5 times), the weight adjustment mechanism is triggered; Correlation threshold: When the correlation coefficient between environmental anomalies and behavioral anomalies exceeds the preset correlation threshold (default value is 0.6), the weight adjustment mechanism is triggered. A correlation threshold of 0.6 indicates a moderate positive correlation, which can be adjusted according to the actual data distribution. The stronger the correlation, the greater the weight adjustment. Weight adjustment will only be performed when both of the above triggering conditions are met simultaneously.

[0042] In this embodiment, the system has a preset dynamic weight function library, which contains a variety of weight calculation functions that can be selected and used according to actual needs. This embodiment provides the following typical functions: Behavioral factor weight = base weight × (1 + coupling frequency / baseline frequency × correlation coefficient); Dynamic weight of environmental factors = 1 - Dynamic weight of behavioral factors; In this embodiment, the initial value of the basic weight is set to 0.5 (i.e., environmental factors and behavioral factors each account for 0.5), the baseline frequency is taken as the frequency threshold (5 times), the coupling frequency is the actual statistical value, and the correlation coefficient is the calculated Pearson coefficient. To prevent excessive weighting of any one factor, the system sets upper and lower limits for weights. The upper limit for behavioral factors is 0.8, and the lower limit is 0.3. Correspondingly, the lower limit for environmental factors is 0.2, and the upper limit is 0.7. When the calculation result exceeds the range, the boundary value is taken.

[0043] Here is an example of weight calculation: Assuming that at a certain monitoring point, the frequency of coupled risk events was 8 times in the past 24 hours (exceeding the threshold of 5 times), and the correlation coefficient between environmental anomalies and behavioral anomalies was 0.75 (exceeding the threshold of 0.6), with the baseline frequency set at 5 times, then the dynamic weight of the behavioral factor = 0.5 × (1 + 8 / 5 × 0.75) = 0.5 × (1 + 1.6 × 0.75) = 0.5 × (1 + 1.2) = 0.5 × 2.2 = 1.1. Since the calculated weight of the behavioral factor exceeds the upper limit of 0.8, the upper limit of 0.8 is taken. Therefore, the dynamic weight of the environmental factor = 1 - 0.8 = 0.2. This result indicates that at this monitoring point in the current period, the contribution of behavioral anomalies to fire risk is much higher than that of environmental anomalies. Therefore, behavioral factors should be given a higher weight in subsequent risk fusion calculations.

[0044] In one embodiment of the present invention, the multimodal fusion publicity module is connected to the comprehensive risk assessment and decision-making module, and includes the following sub-modules: The modal collaborative control submodule is used to receive warning response commands and extract the warning theme, warning intensity and output timing parameters contained therein; The voice broadcast submodule is used to broadcast the corresponding voice warning signal according to the volume level in the warning intensity parameter. The voice warning signal includes at least one or more of the following: fire prevention warning, legal and regulatory prompts, and emergency guidance. The graphic display submodule includes an LED display screen and a driving circuit, which is used to dynamically display the corresponding graphic warning signal according to the display brightness level and flashing frequency in the warning intensity parameters; The timing synchronization control module is used to control the output order and timing ratio of voice warning signals and graphic warning signals according to the output timing parameters.

[0045] The multimodal integrated publicity module also includes the following sub-modules: The environmental perception submodule extracts ambient light intensity and ambient noise decibel values ​​from environmental parameters. The output parameter dynamic adjustment submodule is connected to the environmental perception submodule and the modal collaborative control submodule, respectively. It is used to automatically adjust the display brightness of the LED display screen according to the ambient light intensity and automatically adjust the output volume of the voice broadcast submodule according to the ambient noise decibel value.

[0046] It should be noted that in this embodiment, the modal cooperative control submodule uses an embedded microcontroller as the core processing unit, and is connected to the comprehensive risk assessment and decision-making module via an RS485 bus or wireless communication. When a warning response command is received, the command is first verified and parsed, and the following parameters are extracted: Warning topic parameters: These include the audio file name, LED display text content, and display image name, which are used to determine the specific content to be played. Warning intensity parameters include voice broadcast volume level (adjustable from 1 to 10 levels), LED display brightness level (adjustable from 1 to 10 levels), and LED flashing frequency (adjustable from 0 to 10 times / second), which are used to control the physical intensity of the propaganda output. Output timing parameters include output modes (such as voice priority mode, image and text priority mode, audio-visual synchronization mode, alternating reinforcement mode), duration of each mode, alternation period, total playback duration, etc., which are used to control the order and timing ratio of voice and image and text output. After parsing, the modal collaborative control submodule sends voice-related parameters to the voice broadcasting submodule and image-text related parameters to the image-text display submodule. At the same time, it starts an internal timer to coordinate the output rhythm of the two according to the timing parameters.

[0047] To more clearly illustrate the data coupling relationship between the sub-modules, the following describes the complete workflow using a typical application scenario as an example: Scenario: The integrated risk assessment and decision-making module generates a high-risk warning response instruction, requiring the adoption of an alternating reinforcement mode to enhance the publicity effect; The modal cooperative control submodule receives the warning response command and parses out the following parameters: Voice message: Approximately 15 seconds of warning message; Text and image content: The text reads "Red alert for fire risk, open fires strictly prohibited," along with corresponding warning images; Volume levels: 8 (levels 1-10); Brightness level: 9 (levels 1-10); Flashing frequency: 5 times / second; Output mode: Alternating reinforcement mode; Voice message duration: 5 seconds; Duration of one image / text segment: 3 seconds; Number of alternation cycles: 3 cycles.

[0048] The modal collaborative control submodule sends voice parameters to the voice broadcasting submodule, sends graphic and text parameters to the graphic and text display submodule, and initiates timing synchronization control; The timing synchronization control module initiates the alternating reinforcement mode state machine: Cycle 1: The voice broadcast submodule plays voice for 5 seconds, while the image and text display submodule remains silent; after 5 seconds, the voice is paused, the image and text display submodule is started, and the image and text are displayed for 3 seconds; Cycle 2: Pause the text and image display, then restart the voice broadcast submodule and play for 5 seconds; after 5 seconds, pause the voice broadcast, then restart the text and image display for 3 seconds. Cycle 3: Repeat the above process; After completing 3 cycles, stop all outputs and return to standby mode.

[0049] During voice playback, the voice broadcast submodule broadcasts warning voice messages at volume level 8. During graphic display, the graphic display submodule displays red warning graphics at brightness level 9 and a flashing frequency of 5 times / second. The voice and graphic outputs alternately, creating a double-enhanced warning effect that improves pedestrians' attention to and memory of the warning information.

[0050] In one embodiment of the present invention, the remote interaction and control module is connected to the front-end perception and acquisition module, the behavior feature analysis module, and the multimodal fusion publicity module, and includes the following sub-modules: The data aggregation and processing submodule is connected to the front-end perception and acquisition module, the behavior feature analysis module, and the comprehensive risk assessment and decision-making module, respectively, and is used to receive and aggregate video stream data, behavior feature tags, and comprehensive risk levels. The 3D visualization display submodule is used to display the location distribution, equipment operation status, real-time video footage and comprehensive risk level of each forest fire monitoring point on a 3D map in real time, based on a 3D geographic information system engine. The equipment status monitoring submodule is used to monitor the online status, remaining power, storage space, and fault information of each front-end device in real time.

[0051] It should be noted that in this embodiment, the 3D visualization display submodule uses Cesium.js or Three.js as the 3D map rendering engine, combined with WebGIS technology to realize the 3D map display on the browser. On the 3D map, each forest fire monitoring point is marked with a 3D model icon. The icon style is distinguished according to the equipment type and operating status, as shown in Table 4: Table 4: Illustration of Icon Styles for 3D Models of Forest Fire Prevention Monitoring Points Clicking on any monitoring point icon will bring up an information pop-up window displaying the following information: Monitoring point ID, location coordinates, real-time comprehensive risk level, current environmental data (temperature, humidity, wind speed), time and type of the most recent event, and equipment status summary (battery level, storage status, signal strength, etc.). In addition, the 3D visualization display submodule supports direct viewing of real-time video footage on the 3D map. When the administrator clicks the monitoring point icon and selects the "View Video" option, the system pulls the real-time video stream from the edge computing submodule and pops up a video window in the 3D scene, which is displayed above the map.

[0052] In one embodiment of the present invention, the remote interaction and control module further includes: The remote broadcasting submodule allows administrators to initiate real-time remote broadcasts via microphone. When remote broadcasting is initiated, the multimodal fusion publicity module automatically pauses the preset voice broadcast and prioritizes playing the administrator's real-time voice. The remote update submodule for publicity content is used to send instructions to update publicity content to designated or grouped forest fire prevention monitoring points via a communication network, including changing voice files, modifying LED text content, and setting playback modes.

[0053] The remote interaction and control module also includes a one-click call submodule, which includes: The call terminal submodule, integrated with the multimodal fusion publicity module, includes a physical call button, microphone, and speaker, for pedestrians to initiate call requests with on-duty management personnel. The call routing and allocation submodule is used to receive call requests from pedestrians and allocate the call requests to the call terminal of the duty manager; The call establishment and control submodule is connected to the call routing and allocation submodule and the remote calling submodule, respectively. It is used to establish a two-way voice call link after the call request is answered, and at the same time automatically retrieve the real-time video screen of the monitoring point and display it synchronously to the manager's terminal.

[0054] It should be noted that, in this embodiment, the remote announcement submodule consists of the following parts: Management side: Computers or mobile terminals deployed in the monitoring center, equipped with microphones, speakers and audio acquisition devices, and installed with remote management platform clients or accessed via a web browser; Cloud signaling server: responsible for establishing and managing real-time communication links between the management terminal and the front-end devices, and handling the forwarding of call requests and session control; Front-end device: An edge computing submodule deployed at forest fire monitoring points, with built-in audio codec unit and power amplifier circuit, sharing a speaker with the voice broadcast submodule. The front-end device maintains an audio playback task queue, sorted by the following priority: Priority 1 (highest): Remote broadcasting; Priority 2: Real-time alert voice (such as voice triggered by high-risk levels); Priority 3: Pre-set promotional audio (such as routine fire prevention announcements); When remote voice communication is initiated, the front-end device pauses any currently playing audio task, saves the playback state to memory, and then immediately switches to remote voice communication audio output. After the voice communication ends, the front-end device resumes the paused task based on the saved state (such as continuing playback from the breakpoint). If no state is saved, it returns to standby.

[0055] The remote update submodule for promotional content consists of the following parts: Content Management Backend: A web management interface deployed on a cloud server, allowing administrators to upload, edit, and manage promotional content. It supports functions such as voice file management, text content editing, image material management, and playback strategy configuration. Content distribution service: responsible for sending the content and instructions configured by the administrator to the front-end devices, handling device confirmation and feedback, and supporting three delivery modes: unicast, multicast and broadcast; Front-end device: The edge computing submodule deployed at forest fire monitoring points is responsible for receiving update instructions, downloading new content, updating local storage, and restarting related services, and is used to report update results.

[0056] Furthermore, it should be noted that the present invention can be provided as a method, apparatus, or computer program product. Therefore, embodiments of the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, embodiments of the present invention can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0057] Embodiments of the present invention are described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0058] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The functions specified in one or more boxes. These computer program instructions may also be loaded onto a computer or other programmable data processing terminal equipment to cause a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0059] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. The terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device 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 terminal device. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0060] Finally, it should be noted that the above description represents a preferred embodiment of the present invention. It should be pointed out that although preferred embodiments have been described, those skilled in the art, once they understand the basic inventive concept of the present invention, can make various improvements and modifications without departing from the principles described herein. These improvements and modifications should also be considered within the scope of protection of the present invention. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the embodiments of the present invention.

Claims

1. An intelligent forest fire prevention publicity system based on video surveillance, characterized in that, include: The front-end sensing and acquisition module is deployed at forest fire monitoring points to collect video stream data and environmental data of the monitoring area in real time. The behavior feature analysis module is used to identify pedestrian and vehicle targets in the video stream data, analyze their current location and intention orientation, and assign behavior feature labels based on the intention orientation; The integrated risk assessment and decision-making module is used to integrate and analyze the behavioral feature tags and the environmental data to generate an integrated risk level, and to match the corresponding warning theme, warning intensity and output sequence from the preset publicity strategy library according to the integrated risk level to generate a warning response instruction; The multimodal fusion publicity module is used to respond to the warning response command and play voice warning signals and dynamic graphic warning signals at the forest fire prevention monitoring point; The remote interaction and control module is used to receive and display the video stream data, the behavioral feature tags and the comprehensive risk level, and send the publicity content update instructions and remote broadcast instructions to the multimodal fusion publicity module. It also supports two-way communication between pedestrians and managers.

2. The intelligent forest fire prevention publicity system based on video surveillance according to claim 1, characterized in that: The front-end sensing and acquisition module Includes the following sub-modules: The video acquisition submodule is deployed on the top of the pole at the forest fire monitoring point and is used to collect video stream data of the monitoring area in real time. The environmental monitoring submodule, integrated with the video acquisition submodule on the same pole, is used to collect environmental data of the monitoring area, including temperature, humidity and wind speed. An infrared sensing submodule, deployed in conjunction with the video acquisition submodule, is used to detect whether pedestrians or vehicles are passing through the monitoring area. The edge computing submodule, connected to the video acquisition submodule and the infrared sensing submodule, is used to perform front-end preprocessing on the video stream data, including motion detection and human / vehicle recognition, and triggers video stream data storage and uploading only when a pedestrian or vehicle is detected.

3. The intelligent forest fire prevention publicity system based on video surveillance according to claim 1, characterized in that: The behavior feature analysis module is connected to the front-end perception and acquisition module, and includes the following sub-modules: The target detection and tracking submodule is connected to the front-end perception and acquisition module. It is used to identify pedestrian targets and vehicle targets from the video stream data, track their real-time position, direction of movement and speed of movement, and generate motion trajectory data containing the target ID. The dwell time analysis submodule is used to calculate the dwell time of the pedestrian and vehicle targets within the monitoring area; The trajectory pattern recognition submodule is used to analyze the target's motion trajectory pattern, including path direction, number of hovering times, and entry / exit direction; The intent orientation determination submodule is used to fuse the dwell time with the motion trajectory pattern features to generate corresponding behavioral feature labels.

4. The intelligent forest fire prevention publicity system based on video surveillance according to claim 1, characterized in that: The comprehensive risk assessment and decision-making module is connected to the front-end perception and acquisition module and the behavioral feature analysis module, respectively, and includes a spatiotemporally coupled risk identification submodule, which is used for: The received environmental data and behavioral feature labels are aligned with timestamps and spatial coordinates to generate a spatiotemporally synchronized environment-behavior joint dataset. Environmental anomaly features and behavioral anomaly features are extracted from the environment-behavior joint dataset. The environmental anomaly features include at least one of temperature anomaly, humidity anomaly, and wind speed anomaly. The behavioral anomaly features include at least one of abnormal loitering, wandering trajectory, and fire use behavior including smoking. The built-in coupling risk discrimination model identifies the correlation between environmental anomalies and behavioral anomalies that occur simultaneously within the same spatiotemporal window, determines whether the two constitute a coupling risk, and encapsulates the environmental anomalies and behavioral anomalies that constitute a coupling risk into a coupling risk event pair. The coupling risk event pair is used to input into a risk fusion calculation model to generate the comprehensive risk level.

5. The intelligent forest fire prevention publicity system based on video surveillance according to claim 4, characterized in that: The comprehensive risk assessment and decision-making module also includes a dynamic weight adjustment submodule, which is used for: The frequency of occurrence of the coupled risk events within a continuous spatiotemporal window is statistically analyzed, and the correlation coefficient between environmental anomalies and behavioral anomalies is calculated. Based on the built-in dynamic weight function library, the dynamic weight coefficients of environmental factors and behavioral factors within the current spatiotemporal window are calculated according to the occurrence frequency and the correlation coefficient. When the occurrence frequency exceeds a preset frequency threshold and the correlation coefficient exceeds a preset correlation threshold, the dynamic weight coefficient of the behavioral factor is automatically increased according to a preset increment function. The dynamic weighting coefficients are output to the risk fusion calculation model, which is used to perform weighted fusion of the environmental anomaly features and the behavioral anomaly features when generating the comprehensive risk level.

6. The intelligent forest fire prevention publicity system based on video surveillance according to claim 1, characterized in that: The multimodal integrated publicity module is connected to the comprehensive risk assessment and decision-making module, and includes the following sub-modules: The modal collaborative control submodule is used to receive the warning response command and extract the warning theme, warning intensity and output timing parameters contained therein; The voice broadcast submodule is used to broadcast the corresponding voice warning signal according to the volume level in the warning intensity parameter. The voice warning signal includes at least one or more of the following: fire prevention warning, legal and regulatory prompts, and emergency guidance. The graphic display submodule includes an LED display screen and a driving circuit, which is used to dynamically display the corresponding graphic warning signal according to the display brightness level and flashing frequency in the warning intensity parameters; The timing synchronization control module is used to control the output order and time ratio of the voice warning signal and the graphic warning signal according to the output timing parameters.

7. The intelligent forest fire prevention publicity system based on video surveillance according to claim 6, characterized in that: The multimodal fusion publicity module also includes the following sub-modules: The environmental perception submodule extracts ambient light intensity and ambient noise decibel values ​​from environmental parameters. The output parameter dynamic adjustment submodule is connected to the environment perception submodule and the modal collaborative control submodule, respectively, and is used to automatically adjust the display brightness of the LED display screen according to the ambient light intensity, and automatically adjust the output volume of the voice broadcast submodule according to the ambient noise decibel value.

8. The intelligent forest fire prevention publicity system based on video surveillance according to claim 1, characterized in that: The remote interaction and control module is connected to the front-end perception and acquisition module, the behavior feature analysis module, and the multimodal fusion publicity module, and includes the following sub-modules: The data aggregation and processing submodule is connected to the front-end perception and acquisition module, the behavior feature analysis module, and the comprehensive risk assessment and decision-making module, respectively, and is used to receive and aggregate the video stream data, the behavior feature tags, and the comprehensive risk level; The 3D visualization display submodule is used to display the location distribution, equipment operation status, real-time video footage and comprehensive risk level of each forest fire monitoring point on a 3D map in real time, based on a 3D geographic information system engine. The equipment status monitoring submodule is used to monitor the online status, remaining power, storage space, and fault information of each front-end device in real time.

9. The intelligent forest fire prevention publicity system based on video surveillance according to claim 1, characterized in that: The remote interaction and control module also includes: The remote broadcasting submodule is used to allow the administrator to initiate real-time remote broadcasting through a microphone. When remote broadcasting is started, the multimodal fusion publicity module automatically pauses the preset voice broadcast and prioritizes playing the administrator's real-time voice. The remote update submodule for publicity content is used to send instructions to update publicity content to designated or grouped forest fire prevention monitoring points via a communication network, including changing voice files, modifying LED text content, and setting playback modes.

10. The intelligent forest fire prevention publicity system based on video surveillance according to claim 1, characterized in that: The remote interaction and control module further includes a one-click call submodule, which includes: The call terminal submodule, integrated with the multimodal fusion publicity module, includes a physical call button, microphone, and speaker, for pedestrians to initiate a call request with the duty manager. The call routing and allocation submodule is used to receive call requests from pedestrians and allocate the call requests to the call terminal of the duty manager; The call establishment and control submodule is connected to the call routing and allocation submodule and the remote calling submodule, respectively. It is used to establish a two-way voice call link after the call request is answered, and at the same time automatically retrieve the real-time video screen of the monitoring point and display it synchronously to the manager's terminal.

Images